Sources and References

Primary textbook — Malone PM, Malone MJ, Park SK. Drug Information: A Guide for Pharmacists, 6th ed. McGraw-Hill, 2018. Available through AccessPharmacy.

Supplementary texts — Guyatt G, Rennie D, Meade MO, Cook DJ. Users’ Guide to the Medical Literature: A Manual for Evidence-Based Clinical Practice, 3rd ed. McGraw-Hill, 2015. Hulley SB, Cummings SR, Browner WS, Grady DG, Newman TB. Designing Clinical Research, 4th ed. Lippincott Williams & Wilkins, 2013. Jekel JF, Katz DL, Elmore JG, Wild DMG. Epidemiology, Biostatistics, and Preventive Medicine, 4th ed. Saunders, 2014.

Online resources — PubMed/MEDLINE (https://pubmed.ncbi.nlm.nih.gov); Cochrane Library (https://www.cochranelibrary.com); Lexicomp (https://online.lexi.com); Micromedex (https://www.micromedexsolutions.com); Clinical Pharmacology (https://www.clinicalpharmacology.com); UpToDate (https://www.uptodate.com); RxTx/CPS (Canadian Pharmacists Association); Health Canada Drug Product Database; Clinicaltrials.gov; GRADE Working Group resources (https://www.gradeworkinggroup.org).

Chapter 1: The Nature of Drug Information — Types, Sources, and the Information Hierarchy

Introduction to Drug Information Practice

Drug information is the cornerstone of evidence-based pharmacy practice. Every clinical decision a pharmacist makes — whether recommending a medication, detecting an interaction, counselling a patient, or advising a prescriber — is grounded in information: information about the drug’s pharmacology and pharmacokinetics, information about the disease and the patient’s clinical context, information about the quality and applicability of the evidence supporting the recommendation. The pharmacist who cannot efficiently locate, critically evaluate, and accurately communicate drug information cannot practice at the standard that patients deserve.

Drug information practice encompasses much more than knowing what resources exist and how to look things up. It involves understanding the architecture of the biomedical literature — how evidence is generated, published, organized, and indexed — so that one can construct efficient search strategies and retrieve the most relevant and highest-quality evidence for a given clinical question. It involves critically appraising the methodology of primary research studies — understanding what features of study design protect against bias and what limitations restrict the generalizability of findings. It involves synthesizing and communicating evidence in a manner that is tailored to the question and the audience — a brief verbal summary for a colleague on the hospital ward, a structured written response for a formulary committee, a plain-language explanation for a patient making a treatment decision. And it involves recognizing the role of emerging tools, including artificial intelligence, in drug information practice — their potential and their substantial current limitations.

The field of drug information as a formal pharmacist specialty emerged in the 1960s with the establishment of the first drug information centers — initially at the University of Kentucky in 1962, rapidly spreading to academic medical centers across North America. Drug information centers provided pharmacists and physicians with authoritative responses to clinical questions that were beyond the resources of practitioners in individual hospitals or communities. Today, while dedicated drug information centers still operate (the Canadian Drug Information Centre, operated by the Canadian Pharmacists Association, is one example), the widespread availability of electronic databases and online resources has democratized access to drug information, making drug information skills a competency required of every practicing pharmacist rather than a specialized service provided by a few centers.

Primary, Secondary, and Tertiary Literature

The biomedical literature is conventionally organized into three levels based on the degree of processing and synthesis the information has undergone. Understanding this hierarchy is fundamental to designing efficient search strategies and selecting the most appropriate source for a given clinical question.

The optimal search strategy for a drug information question often begins with a tertiary source to obtain a quick overview and identify known clinical considerations, then moves to secondary databases to identify recent primary literature that may have updated or refined the tertiary source’s information, and finally involves critical appraisal of the most relevant primary studies. For some questions — particularly those in rapidly evolving areas such as new drug safety signals, emerging infectious diseases, or recently approved therapies — the most current evidence may exist only in the primary literature, and direct primary literature searching is essential.

Chapter 2: Key Drug Information Resources — Tertiary Tools and Their Applications

Drug Compendia and Monograph Resources

The most frequently consulted drug information resources in daily pharmacy practice are drug monographs — structured summaries of a drug’s clinical pharmacology, approved indications, dosing, contraindications, warnings, adverse effects, drug interactions, and other clinical information. In Canada, the most important sources of authoritative drug monograph information are the product monographs approved by Health Canada (which are legally binding documents describing the conditions of drug authorization), the Compendium of Pharmaceuticals and Specialties (CPS, now available electronically as RxTx through the Canadian Pharmacists Association), Lexicomp, and Micromedex.

Health Canada product monographs are the gold standard for regulatory-approved drug information in Canada. A product monograph is a scientific document prepared by the manufacturer and reviewed and approved by Health Canada as part of the drug approval process; it describes the drug’s chemical and pharmaceutical characteristics, pharmacokinetics, pharmacodynamics, clinical efficacy (with references to the key clinical trials on which approval was based), safety information, and instructions for use. Product monographs in Canada are organized into three sections: Part I (Health Professional Information — the prescribing information), Part II (Scientific Information — detailed pharmacological and clinical data), and Part III (Consumer Information — a plain-language summary). Product monographs are authoritative for approved indications but do not necessarily reflect the most current evidence from post-marketing studies; they may also not address off-label uses that are nonetheless supported by evidence.

Lexicomp is a comprehensive, continuously updated drug information database widely used in Canadian and North American hospital and community pharmacy practice. It provides detailed drug monographs, drug interaction analysis, dosing databases (including specialized pediatric and renal dosing references), pharmacogenomics information, patient education materials, and natural health product information. Its drug interaction analysis module checks for interactions between two or more drugs entered by the user, classifying interactions by severity and quality of evidence and providing mechanistic explanations and management recommendations. The interaction database is one of the most comprehensive available but is not infallible: like all drug interaction databases, it is subject to over-alerting (flagging interactions of limited clinical significance) and under-alerting (missing rare but serious interactions with limited published evidence), and clinical judgment is required in interpreting its alerts.

Micromedex (IBM Watson Health) is another major drug information platform widely available in hospital settings. Its DRUGDEX component provides evidence-based monographs with explicit grading of the evidence for each indication. The Thomson Clinical Pharmacology module provides drug monographs, drug interaction analysis, and drug comparison tools. Micromedex is particularly valued for its clinical pharmacokinetics calculator, drug identification database (IDENTIDEX), and comprehensive toxicology information (POISINDEX), which is an essential resource for pharmacists working in poison control contexts. In Canadian hospitals, Micromedex and Lexicomp are often available simultaneously, and pharmacists use them complementarily — consulting multiple resources for complex drug information questions and noting concordance or discordance between sources.

Clinical Practice Guidelines

Clinical practice guidelines (CPGs) are systematically developed statements intended to assist practitioner and patient decisions about appropriate healthcare for specific clinical circumstances. In pharmacy practice, CPGs are among the most important tertiary resources because they synthesize the evidence on a given condition and provide explicit, graded recommendations for drug therapy — identifying which drugs are first-line versus second-line versus third-line, specifying doses and monitoring parameters, and flagging special considerations for specific patient populations.

Critically evaluating a clinical practice guideline requires an understanding of how guidelines are developed and the potential for bias to enter the process. The highest-quality CPGs use systematic literature search methods to identify relevant evidence, appraise the quality of each study using validated tools (such as the Cochrane Risk of Bias tool for RCTs or the Newcastle-Ottawa Scale for observational studies), synthesize the evidence using frameworks such as GRADE (Grading of Recommendations, Assessment, Development and Evaluations), and produce recommendations with explicit ratings of the strength of the recommendation and the quality of the underlying evidence. A GRADE “strong recommendation” based on “high-quality evidence” (typically from multiple well-designed RCTs with consistent results) carries much more evidential weight than a “conditional recommendation” based on “low-quality evidence” (typically from observational studies or RCTs with methodological limitations).

The GRADE system rates evidence quality using four levels. High quality evidence: further research is very unlikely to change confidence in the effect estimate (typically based on consistent, well-designed RCTs). Moderate quality evidence: further research is likely to have an important impact on confidence (typically based on RCTs with important limitations, or consistent observational studies). Low quality evidence: further research is very likely to change confidence (typically based on observational studies, or RCTs with major limitations). Very low quality evidence: any estimate of effect is very uncertain (typically based on case series, expert opinion, or indirect evidence). The GRADE framework also rates recommendation strength as strong or conditional (weak), based on the tradeoff between benefits and harms, the quality of evidence, patient values and preferences, and resource implications.

Natural Health Product Resources

Natural health products (NHPs) — herbal remedies, vitamins and minerals, homeopathic preparations, traditional Chinese medicines, and other non-prescription products of natural origin — present particular challenges for drug information practice. The evidence base for NHPs is far less developed than for pharmaceutical drugs: most have not been studied in rigorous randomized controlled trials for their proposed therapeutic claims, their standardization of active ingredients can be inconsistent between products and manufacturers, and they are subject to a lighter regulatory burden than prescription drugs under Canada’s Natural Health Products Regulations (part of the Food and Drugs Act).

In Canada, NHPs are regulated by Health Canada’s Natural and Non-prescription Health Products Directorate (NNHPD). To be sold legally in Canada, an NHP must have a product licence from Health Canada (indicated by a Natural Product Number [NPN] or Homeopathic Medicine Number [DIN-HM] on the label), demonstrating that it meets standards for safety, efficacy, and quality based on a review of the submitted evidence. However, the evidence standard for NHP licensing is lower than for prescription drugs, and the “efficacy” claims on licensed NHP labels may be supported by traditional use evidence rather than clinical trial data. This distinction is critically important for pharmacists counselling patients: a Health Canada NPN license does not mean a product has been shown in rigorous clinical trials to be effective for its claimed use — it means the product meets minimum safety and quality standards and that its efficacy claims are supported by the evidence level required for NHP licensing.

Key NHP drug information resources include Natural Medicines (formerly Natural Medicines Comprehensive Database), a subscription database that provides systematic reviews of the evidence for NHPs with explicit evidence ratings; the Natural Standard database; and the Memorial Sloan Kettering Cancer Center’s “About Herbs” resource, which is particularly valuable for oncology patients asking about herbal products during cancer treatment. The issue of NHP-drug interactions is of substantial clinical importance: St. John’s Wort (Hypericum perforatum), used for mild depression, is a potent inducer of CYP3A4, P-glycoprotein, and CYP2C9, substantially reducing the plasma concentrations of cyclosporine, antiretrovirals, warfarin, oral contraceptives, digoxin, and many other drugs to potentially sub-therapeutic levels. Ginkgo biloba has antiplatelet effects and may increase bleeding risk when combined with anticoagulants or antiplatelet agents. Valerian, kava, and melatonin have additive CNS depressant effects with sedative drugs.

Chapter 3: Searching the Primary Literature — Databases and Search Strategy

MEDLINE/PubMed — The Primary Biomedical Literature Database

MEDLINE, produced by the United States National Library of Medicine (NLM) and freely accessible through PubMed, is the most comprehensive and widely used bibliographic database of biomedical literature in the world. It contains over 35 million citations from more than 5,000 biomedical journals worldwide, dating back to the 1940s for many journals and more completely from the 1950s forward. For pharmacy students and practicing pharmacists, PubMed is the essential primary literature search tool.

Each citation in MEDLINE is indexed with Medical Subject Headings (MeSH) — a controlled, hierarchical vocabulary of biomedical terms maintained and updated annually by the NLM. MeSH indexing allows searchers to retrieve records about a concept regardless of what terminology individual authors used, because all articles about a topic are indexed with the same standardized MeSH terms. For example, a search for the MeSH term “Myocardial Infarction” will retrieve articles that use any of the synonyms for heart attack — myocardial infarction, MI, AMI, acute myocardial infarction, heart attack — because all of these articles have been indexed with the MeSH heading “Myocardial Infarction” by NLM indexers. Understanding MeSH structure, including the concept of MeSH trees (which organize terms hierarchically from broad to specific) and the ability to “explode” a MeSH term to retrieve all narrower terms within its tree, is essential for conducting comprehensive systematic literature searches.

A well-constructed PubMed search for a specific clinical drug information question typically combines MeSH terms and keyword searching to ensure both sensitivity (not missing relevant articles) and specificity (not retrieving an unmanageable number of irrelevant articles). The basic building blocks of a database search are Boolean operators: AND (narrows the search by requiring both terms), OR (broadens the search by retrieving records with either term), and NOT (excludes records with the specified term). A search for pharmacist-led interventions in heart failure might combine: (“Pharmacists” [MeSH] OR “Pharmacy Service, Hospital” [MeSH] OR “pharmacist” [tw]) AND (“Heart Failure” [MeSH]) AND (“Treatment Outcome” [MeSH] OR “medication adherence” [MeSH]). PubMed also offers a “Clinical Queries” filter that limits results to specific study designs (therapy, diagnosis, prognosis, etiology, clinical prediction guides) and can be set for either high sensitivity (retrieving most relevant studies) or high specificity (retrieving mostly relevant studies), a tradeoff that must be balanced based on the question at hand.

EMBASE and International Pharmaceutical Abstracts

EMBASE (Excerpta Medica, produced by Elsevier) is a European biomedical database that complements MEDLINE/PubMed, particularly for pharmacological, pharmacokinetic, and drug safety literature. EMBASE indexes approximately 8,500 journals, with stronger coverage of European publications than MEDLINE, and its thesaurus (EMTREE) provides a controlled vocabulary broadly comparable to MeSH. For comprehensive systematic reviews on drug therapy questions, it is standard practice to search both MEDLINE and EMBASE, as studies identified by searching EMBASE but not MEDLINE (and vice versa) can meaningfully alter the conclusions of a review.

International Pharmaceutical Abstracts (IPA), produced by the American Society of Health-System Pharmacists (ASHP) and available through several database platforms, indexes pharmacy-specific journals and pharmacy conference proceedings that are not indexed in MEDLINE or EMBASE. This makes IPA particularly valuable for identifying practice-focused pharmacy literature — pharmacist interventions, medication use evaluations, drug information practice, pharmacy education research, and drug policy — that may not appear in the general biomedical databases. For questions about pharmacy practice outcomes, pharmaceutical policy, or drug information service utilization, IPA is an important complementary database.

The Cochrane Library provides access to the Cochrane Database of Systematic Reviews (CDSR) — widely regarded as the highest-quality source of systematic reviews on healthcare interventions — as well as Cochrane CENTRAL, a database of controlled trials assembled from MEDLINE, EMBASE, and other sources. Cochrane systematic reviews follow a rigorous, pre-specified protocol and use validated methods for study selection, data extraction, quality assessment (using the Cochrane Risk of Bias tool), and meta-analysis; they are regularly updated as new evidence becomes available. For questions about therapeutic efficacy — which drug works better for a given condition, whether a drug is better than placebo — a current Cochrane systematic review, when one exists, typically provides the most reliable synthesized evidence available.

Constructing a Systematic Search: PICO and the Search Strategy

For clinical drug information questions — particularly those arising in evidence-based patient care — the PICO framework provides a structured approach to formulating the question in a way that can be translated directly into an efficient literature search. PICO stands for Population (who is the patient or patient group?), Intervention (what drug or therapy is being considered?), Comparator (what is being compared — placebo, usual care, another drug?), and Outcome (what patient-important outcomes matter — efficacy endpoints, safety endpoints, quality of life?). Formulating a PICO question before searching not only improves search efficiency but clarifies the scope of the question, making it easier to assess whether retrieved literature actually answers it.

Chapter 4: Study Design in Clinical Research

The Evidence Hierarchy and Study Design Overview

The quality of evidence generated by a clinical study is fundamentally determined by its design. Different study designs have different vulnerabilities to bias — systematic error in the design, conduct, analysis, or interpretation of a study that produces results that deviate from the truth. The evidence hierarchy, popularized by evidence-based medicine proponents in the 1990s and formalized in frameworks such as GRADE, places study designs in a rough order of susceptibility to bias, with randomized controlled trials at the top and case reports and expert opinion at the bottom. Understanding this hierarchy — and, more importantly, understanding why certain designs are more or less susceptible to specific types of bias — is essential for critical appraisal.

The fundamental threat to causal inference in non-randomized studies is confounding: a situation in which an observed association between an exposure (drug) and an outcome is distorted by a third variable (the confounder) that is associated with both the exposure and the outcome independently. For example, if a study finds that patients who take low-dose aspirin have a lower risk of colorectal cancer, this association could be confounded by healthfulness — aspirin users may also be more health-conscious, exercise more, eat more vegetables, and smoke less, and it is these healthy behaviors (not the aspirin) that reduce cancer risk. Confounding by indication is particularly problematic in pharmacoepidemiology: patients who are prescribed a drug (e.g., statins for cardiovascular prevention) differ systematically from those who are not prescribed it in ways (underlying cardiovascular risk, physician access, health-seeking behavior) that are themselves associated with the outcome under study.

Randomized Controlled Trials — Design and Interpretation

The randomized controlled trial is the gold standard experimental design for establishing causal efficacy of medical interventions because randomization is the only method that controls for both known and unknown confounders. By assigning treatment by chance, randomization ensures that — on average, across a large enough sample — the characteristics of participants in each group are balanced, including characteristics that were not measured or are not known to influence the outcome. This allows any difference in outcomes between groups to be attributed causally to the intervention itself, rather than to pre-existing differences between groups.

Blinding is the concealment of treatment assignment from participants, clinicians, outcome assessors, and data analysts during the trial. Double-blinding (both participants and clinicians blinded) reduces multiple sources of bias: participants who know they are receiving an experimental drug may experience placebo effects that inflate the perceived benefit; clinicians who know a patient’s assignment may provide differential care or differential outcome ascertainment; and outcome assessors who know the assignment may unconsciously score outcomes in a direction consistent with their expectations. Blinding is achievable for drug trials using matching placebos that are identical in appearance, taste, and smell to the active drug. For trials comparing two active drugs with different appearances or dosing regimens, a double-dummy design — in which all participants receive both an active drug and a matching placebo for the comparator — can maintain blinding.

The intention-to-treat (ITT) principle specifies that all participants must be analyzed in the group to which they were randomly assigned, regardless of whether they completed the protocol, took the assigned treatment, or received cross-over treatment. The ITT analysis preserves the benefits of randomization — any differences in outcomes between groups remain attributable to treatment assignment rather than to the characteristics of those who completed or adhered to treatment — and produces an estimate of the treatment effect in a real-world setting where treatment deviations and dropouts occur. The per-protocol analysis, which analyzes only participants who adhered to the assigned treatment, can provide an estimate of the efficacy of the treatment under ideal conditions but is susceptible to bias because adherent and non-adherent participants differ in ways that may be associated with outcomes (sicker patients may be more likely to discontinue).

Cohort Studies — Observational Analytic Design

A cohort study assembles a group of individuals who share a common characteristic (usually an exposure, such as use of a particular drug) and follows them over time to observe the incidence of outcomes. Cohort studies can be prospective (participants are enrolled before the outcomes of interest have occurred and followed forward in time) or retrospective (historical records are used to identify an exposed and unexposed cohort and reconstruct their outcomes retrospectively). Prospective cohort studies offer more control over data collection quality and exposure definition but are expensive and require long follow-up for outcomes with long latency periods. Retrospective cohort studies can be conducted rapidly using existing health databases but are limited by the quality and completeness of the data available in those databases.

The key advantage of cohort studies over case-control studies is their ability to estimate incidence rates — the rate at which new outcomes develop in exposed versus unexposed persons — and directly calculate relative risks (RRs). The key limitation is susceptibility to confounding, which cannot be eliminated by design (unlike randomization) but can be addressed analytically through multivariable regression, propensity score methods, instrumental variable analysis, and other techniques. Pharmacoepidemiology — the study of drug use and effects in large populations using administrative health databases — relies primarily on retrospective cohort and case-control designs, using electronic health records, provincial drug benefit claims data, and linked health administrative databases to study drug effectiveness and safety in real-world patient populations.

In Canada, an exceptionally valuable resource for pharmacoepidemiology is the network of provincial health administrative databases — ICES (Institute for Clinical Evaluative Sciences) in Ontario, the Population Data BC platform in British Columbia, and similar databases in other provinces — which link prescription dispensing records, physician claims, hospital records, and vital statistics for populations of millions of patients over decades. Studies using these databases have contributed importantly to understanding the real-world safety and effectiveness of drugs approved based on clinical trial data: for example, cohort studies of dabigatran versus warfarin in Ontario’s elderly population confirmed the bleeding profile differences seen in the RE-LY trial and identified patient subgroups at particularly high or low bleeding risk.

Case-Control Studies

A case-control study begins with the outcome — it identifies individuals who have experienced the outcome of interest (cases) and individuals who have not (controls) — and then looks backward to compare the frequency of exposure between cases and controls. Case-control studies are particularly efficient for studying rare outcomes, because they start with cases rather than having to follow a large population forward in time until a rare event occurs. They are also efficient for outcomes with long latency periods, as the investigator can enroll cases who already have the disease and ask about historical exposures rather than waiting years for outcomes to develop.

The measure of association in a case-control study is the odds ratio (OR) — the odds of exposure among cases divided by the odds of exposure among controls:

where a is exposed cases, b is exposed controls, c is unexposed cases, and d is unexposed controls. When the outcome is rare (typically below 10%), the OR closely approximates the relative risk (RR) from a cohort study — this is called the “rare disease assumption.” For common outcomes, the OR can substantially overestimate the relative risk in both directions.

The critical methodological challenges in case-control studies are appropriate case and control selection. Controls must be selected from the same source population that gave rise to the cases — that is, if a case had developed the disease, they would have been eligible to be a case; if a control had developed the disease, they would have been enrolled as a case. Failure to achieve this condition — for example, using hospitalized controls for a question in which hospitalization itself is related to the exposure — introduces selection bias. Recall bias is a particular concern in case-control studies: cases who have experienced a serious outcome may recall and report exposures (particularly drug exposures) more thoroughly or differently from controls who have not experienced the outcome, leading to differential misclassification of exposure.

Cross-Sectional Studies and Ecological Studies

A cross-sectional study measures both exposure and outcome status simultaneously in a defined population at a single point in time (or over a brief period). Cross-sectional studies are useful for measuring the prevalence of a condition or a behavior, identifying associations between exposures and prevalent outcomes, and generating hypotheses for analytic studies. They cannot establish temporality (whether the exposure preceded the outcome) and therefore cannot establish causality. They are also susceptible to survivorship bias: individuals who had both the exposure and the outcome but did not survive to be enrolled in the study will not be captured.

Ecological studies analyze aggregate data about populations rather than data about individuals. They examine whether populations (countries, provinces, hospitals, time periods) with higher rates of exposure also have higher rates of the outcome. Ecological studies can be useful for generating hypotheses and for studying population-level interventions, but they are subject to the ecological fallacy — the erroneous inference that patterns observed at the group level apply to individuals. The classic example is the observation that countries with higher per capita fat intake have higher rates of breast cancer across countries, which led to the hypothesis that dietary fat causes breast cancer; individual-level cohort studies largely failed to confirm this association, illustrating that the ecological association was driven by confounders at the country level rather than a causal relationship between individual dietary fat intake and cancer risk.

Chapter 5: Systematic Reviews and Meta-Analysis

The Systematic Review as a Tool for Evidence Synthesis

A systematic review is a structured synthesis of the primary literature on a specific question, conducted using explicit, pre-specified, and reproducible methods to identify, select, and critically appraise all relevant evidence. The defining feature of a systematic review — in contrast to a traditional narrative review — is its systematic, transparent methodology: the research question is framed using PICO, the literature search is comprehensive and documented, eligibility criteria for study inclusion are pre-specified, studies are selected and data extracted by two or more independent reviewers with disagreements resolved by consensus or arbitration, the quality of each included study is assessed using a validated tool, and results are synthesized in a structured, transparent manner.

The Cochrane Collaboration, founded in 1993 and named after the Scottish epidemiologist Archie Cochrane who championed evidence-based medicine, produces and maintains the most rigorously conducted systematic reviews in healthcare. Cochrane systematic reviews follow a standardized protocol published in advance of the review to prevent outcome reporting bias, use comprehensive search strategies spanning multiple databases and grey literature sources, apply the Cochrane Risk of Bias tool (RoB2 for parallel RCTs, RoB2 for crossover trials, ROBINS-I for non-randomized studies) to assess study quality, and are updated regularly as new evidence emerges. The Cochrane Library is freely accessible in Canada through a partnership between the Cochrane Collaboration and the Canadian Medical Association (CMA), making it an essential resource for evidence-based pharmacy practice.

Meta-analysis is the statistical technique of combining quantitative results from multiple primary studies into a pooled effect estimate. When studies address the same clinical question with comparable designs, patient populations, interventions, and outcome measures, meta-analysis can produce a more precise estimate of the treatment effect than any individual study — particularly for outcomes that are too rare for a single trial to detect with adequate statistical power. The pooled effect estimate from a well-conducted meta-analysis — whether expressed as a risk ratio, odds ratio, mean difference, or hazard ratio — comes with a confidence interval that reflects the precision of the estimate across all included studies combined.

The key assumption underlying meta-analysis is that the included studies are sufficiently similar to be meaningfully combined — that the heterogeneity in their designs, populations, interventions, and outcome definitions is not so great that their effect estimates cannot be validly pooled. Heterogeneity between studies is assessed statistically using the I² statistic, which estimates the percentage of the observed variation in effect estimates that is due to true heterogeneity rather than sampling error: I² values of 0-25% suggest low heterogeneity, 25-50% moderate, 50-75% substantial, and above 75% considerable heterogeneity. When significant heterogeneity is present, a random effects model — which treats the true treatment effect as varying across studies and estimates the mean of this distribution — is typically preferred over the fixed effect model (which assumes all studies estimate the same true effect), though the random effects pooled estimate is more conservative and its confidence intervals are wider. Sources of heterogeneity should be explored through subgroup analyses and meta-regression.

Forest Plots and Funnel Plots

The forest plot is the standard graphical representation of a meta-analysis result. Each line of the forest plot represents one included study, with a box centered on the study’s point estimate and horizontal lines extending to the study’s confidence interval limits; the size of the box is typically proportional to the study’s weight in the meta-analysis (larger studies and more precise studies receive more weight). The diamond at the bottom of the forest plot represents the pooled meta-analytic estimate, with its width indicating the confidence interval of the pooled result. A vertical dashed line at the null effect (relative risk or odds ratio of 1.0 for ratio measures, or mean difference of 0 for continuous measures) allows visual assessment of which studies and the overall pooled estimate favor the treatment or control.

Reading a forest plot fluently is an essential pharmacist skill. First, note the direction and magnitude of the pooled estimate — does the intervention reduce or increase risk, and by how much? Second, note whether the pooled confidence interval crosses the null — if it does not, the result is statistically significant. Third, assess heterogeneity — do the individual study estimates cluster tightly around the pooled estimate (low heterogeneity) or scatter widely (high heterogeneity, suggesting that the meta-analytic pooling may be misleading)? Fourth, assess publication bias using the funnel plot: a funnel plot displays each study’s effect estimate against a measure of study precision (typically standard error or sample size); in the absence of publication bias, the points should form an inverted funnel shape symmetric around the pooled estimate. Asymmetry — particularly a deficit of small studies on one side of the funnel — suggests publication bias (small studies with negative results are not being published, inflating the pooled effect estimate toward a more favorable direction for the intervention).

Chapter 6: Biostatistics for Drug Information Practice

Measures of Association and Effect Size

Interpreting the results of clinical research requires facility with a core set of statistical measures. These measures appear throughout the clinical literature, in drug information resources, and in clinical practice guidelines, and the pharmacist who cannot interpret them accurately cannot provide reliable evidence-based drug information.

The relative risk (RR), also called the risk ratio, is the ratio of the probability of an event in the exposed (treatment) group to the probability of the event in the unexposed (control) group:

An RR of 1.0 means no difference between groups; RR < 1 means the treatment reduces the risk (protective); RR > 1 means the treatment increases the risk (harmful). The relative risk reduction (RRR) is the complement of the RR: RRR = 1 - RR = (p_control - p_treatment) / p_control. An RRR of 25% means the treatment reduces the event rate by 25% relative to the control.

The absolute risk reduction (ARR) — also called the absolute risk difference — is the arithmetic difference between event rates: ARR = p_control - p_treatment. The number needed to treat (NNT) is the reciprocal of the ARR: NNT = 1/ARR. The NNT represents the number of patients who must receive the treatment (compared to the control) to prevent one additional outcome event. NNTs vary dramatically across clinical contexts and should always be interpreted in the context of the baseline risk, the severity of the outcome being prevented, and the costs and harms of the treatment.

Confidence Intervals and P-Values

The p-value and the confidence interval are the two most commonly reported measures of statistical inference in clinical trial publications, and both are frequently misunderstood — even by physicians and pharmacists who use them routinely. Precise understanding of these concepts is essential for critically appraising the clinical significance of trial results and for communicating evidence accurately to patients and other healthcare providers.

The p-value is the probability of observing a test statistic at least as extreme as the one obtained (or more extreme) under the assumption that the null hypothesis (no treatment effect) is true. A p-value of 0.03 means that if there were truly no difference between treatments, there would be only a 3% probability of obtaining a difference at least as large as the observed difference by chance. It does not mean that there is a 97% probability that the treatment is effective. It does not describe the magnitude of the effect. And it does not tell you whether the effect is clinically meaningful. The conventional threshold of p < 0.05 for statistical significance is arbitrary — Fisher himself never intended it as a rigid decision boundary — and there is growing consensus in the scientific community that binary statistical significance testing should be supplemented or replaced by interval estimation (confidence intervals) and explicit consideration of effect size and clinical importance.

A 95% confidence interval for a risk ratio or odds ratio is the range of values consistent with the observed data that, if the study were repeated many times, would contain the true parameter value 95% of the time. More practically for clinical interpretation: if the 95% CI for an odds ratio does not include 1.0, the result is statistically significant at the 5% level; if it does include 1.0, the result is not statistically significant at that level. The width of the confidence interval reflects the precision of the estimate — wide CIs indicate a small, imprecise study, while narrow CIs indicate a large, precise study. For a pharmacist, the upper and lower bounds of the CI are often as clinically informative as the point estimate: if the CI for an RR of 0.80 (20% relative risk reduction) extends from 0.55 to 1.15, the true effect could plausibly be anywhere from a 45% reduction to a 15% increase in risk — a range of outcomes that would substantially affect clinical decisions.

Sensitivity, Specificity, and Diagnostic Tests

Although PHARM 155 focuses primarily on therapeutic research, pharmacists working in collaborative care settings must also understand the statistics of diagnostic tests — because the results of diagnostic tests inform the clinical context in which drug therapy decisions are made, and because some pharmacy-based clinical services (point-of-care testing for glucose, lipids, INR, and rapid diagnostics) require interpretation of test results.

Sensitivity and specificity characterize the performance of a diagnostic test against a gold standard. Sensitivity is the probability that the test is positive given that the patient truly has the disease: sensitivity = TP / (TP + FN), where TP is true positives and FN is false negatives. A highly sensitive test rarely misses patients who have the disease — it has few false negatives. Specificity is the probability that the test is negative given that the patient truly does not have the disease: specificity = TN / (TN + FP), where TN is true negatives and FP is false positives. A highly specific test rarely misidentifies healthy patients as diseased — it has few false positives. A highly sensitive test is most useful for ruling out a diagnosis (mnemonic: SnNout — Sensitive test, Negative result, rules Out); a highly specific test is most useful for ruling in a diagnosis (SpPin — Specific test, Positive result, rules In).

Chapter 7: Critical Appraisal of Therapeutic Research

Critical Appraisal Frameworks — The User’s Guide Approach

Critical appraisal is the disciplined process of systematically evaluating a research article to assess the validity of its findings, the magnitude of the effect, and the applicability of the results to a specific patient or clinical setting. The “Users’ Guide to the Medical Literature” framework, developed by Gordon Guyatt and the Evidence-Based Medicine Working Group at McMaster University and originally published in JAMA beginning in 1993, provides the foundational approach to critical appraisal that is most widely used in Canadian pharmacy education.

The Users’ Guide approach addresses three fundamental questions for any clinical research article: (1) Are the results valid? — assessing the internal validity of the study (was it designed and conducted in a way that protects against bias?); (2) What are the results? — understanding the direction, magnitude, and precision of the effect estimate; and (3) Can the results help my patient? — assessing the external validity or generalizability of the study (does the study population resemble my patient? Is the intervention available? Do the outcomes measured reflect outcomes that matter to my patient?). These three questions are applied somewhat differently depending on the study type — the specific questions asked about an RCT differ from those applied to a cohort study, a diagnostic test study, a systematic review, or a clinical practice guideline.

For a randomized controlled trial, the validity assessment asks: Was the treatment randomly allocated? Was the allocation concealed? Were patients, clinicians, and outcome assessors blinded? Were all patients accounted for? Were all patients analyzed in the groups to which they were assigned (ITT analysis)? Were the groups similar at baseline? Were important prognostic variables similar at baseline, or was there imbalance that may have introduced confounding despite randomization (particularly important in small trials)? Was the study stopped early? Trials stopped early for benefit consistently show larger treatment effects than later trials of the same drug that ran to completion — a statistical phenomenon related to the observation that data look more extreme at interim analyses due to random variation, inflating the estimated effect size.

Assessing Bias in Observational Studies

For observational studies — cohort studies, case-control studies — the critical appraisal framework focuses on the threats to validity that cannot be addressed by randomization. The Newcastle-Ottawa Scale (NOS) and the ROBINS-I (Risk of Bias in Non-Randomized Studies of Interventions) tool are the most commonly used instruments for assessing the quality of observational studies in systematic reviews. The key domains of bias assessed include selection bias (are the exposed and unexposed groups comparable at baseline?), information bias (are exposures and outcomes measured accurately and similarly in all groups?), and confounding (have important confounding variables been identified and controlled for in the analysis?).

Confounding by indication is the most pervasive and challenging source of bias in pharmacoepidemiology. When drug prescribing is driven by patient characteristics — sicker patients receive more aggressive drug therapy — comparisons between drug users and non-users are inherently confounded by the severity of the underlying condition. For example, studies comparing outcomes between patients treated with intensive versus standard lipid-lowering therapy may be confounded because physicians who prescribe more intensive therapy may also be more vigilant about other cardiovascular risk factors, resulting in better outcomes for reasons unrelated to the lipid-lowering intensity itself. Propensity score methods — statistical techniques that estimate the probability of receiving the treatment based on observed baseline characteristics, then use this probability to create balanced comparison groups — are increasingly used in pharmacoepidemiology to address confounding by indication, but they can only control for measured confounders; unmeasured confounding remains a fundamental limitation of observational research.

Drug Interaction Critical Appraisal

Drug interaction literature presents specific critical appraisal challenges. The literature on drug interactions is heterogeneous in quality: it includes formal pharmacokinetic drug interaction studies (clinical Phase I trials deliberately designed to assess the effect of one drug on the pharmacokinetics of another — these provide the most reliable quantitative data on the magnitude of an interaction), case reports and case series (which identify the clinical manifestation of interactions but cannot establish causality or estimate incidence), observational database studies using administrative health data (which can estimate the frequency of clinical consequences of interactions in real-world populations), and in vitro studies (which provide mechanistic data on CYP enzyme inhibition or induction but may not predict the magnitude of in vivo interactions because they do not account for drug absorption, distribution, and the fraction of the drug metabolized by the affected enzyme).

When evaluating a drug interaction report, the key questions are: What is the proposed mechanism (pharmacokinetic vs. pharmacodynamic)? How strong is the evidence — is it based on a formal PK study showing a significant magnitude of AUC change, or only on a case report? What is the magnitude of the interaction — a 10% increase in AUC of the affected drug is typically not clinically significant, while a 3-fold increase (200% increase) likely is? Are there clinical reports of harm — do case reports or pharmacovigilance data show actual adverse outcomes in patients taking these drugs together? Can the interaction be managed (dose adjustment, timing separation, monitoring) without avoiding the combination altogether? The clinical significance of a drug interaction cannot be determined by the existence of the interaction alone; it requires integration of mechanism, magnitude, patient vulnerability (renal or hepatic impairment, narrow therapeutic index drugs, elderly or pediatric patients), and available management options.

Chapter 8: Pharmacoeconomics and Health Technology Assessment

Principles of Pharmacoeconomics

Pharmacoeconomics is the discipline that applies economic methods to evaluate the costs and outcomes of drug therapies, providing information to support resource allocation decisions in healthcare. As drug spending continues to increase — driven by biologic therapies, specialty pharmaceuticals, and an aging population with more complex drug needs — the question of whether a new drug therapy represents good value for the resources invested has become central to drug policy and formulary decisions in Canada and internationally.

Health Technology Assessment in Canada

The Canadian Drug Agency (CDA, formerly CADTH — the Canadian Agency for Drugs and Technologies in Health) is the national body responsible for health technology assessment of drugs and medical devices in Canada. CDA conducts Common Drug Reviews (CDRs) for new drugs seeking reimbursement on public drug formularies and Pharmacoeconomic Reviews that assess the cost-effectiveness of new drugs relative to existing therapies. CDA’s reimbursement recommendations — “recommend for listing,” “recommend for listing with conditions,” or “do not recommend for listing” — are non-binding on provincial drug plans but are highly influential in formulary decision-making across Canada.

A CDA Common Drug Review involves both a clinical review (systematic evaluation of the clinical evidence submitted by the manufacturer, supplemented by independent literature searching) and a pharmacoeconomic review (evaluation of the manufacturer’s submitted pharmacoeconomic model, typically a cost-utility analysis, and its assumptions). CDA’s pharmacoeconomic review assesses whether the model is methodologically appropriate, whether the clinical inputs (transition probabilities, utility weights, adverse event rates) are supported by the submitted evidence, and whether the model’s conclusions are robust to sensitivity analysis — particularly to variation in assumptions that have high uncertainty. If the manufacturer’s submitted ICER exceeds approximately $50,000 CAD per QALY (the informal willingness-to-pay threshold used in many Canadian jurisdictions, though this varies and is not officially set), the drug is unlikely to receive a positive recommendation unless the condition is serious, no alternatives exist, or the uncertainty in the ICER estimates is very high.

Chapter 9: Artificial Intelligence and Drug Information

AI in Drug Information Retrieval and Synthesis

The emergence of large language model (LLM)-based AI tools — ChatGPT, Google Gemini, Claude, and others — has generated significant interest and concern in healthcare, including in pharmacy and drug information practice. These tools are capable of generating fluent, apparently authoritative responses to drug information questions, and pharmacists and patients are increasingly using them, making it essential for pharmacy students to understand both what these tools can do and what they cannot reliably do.

Large language models are trained on massive corpora of text — including substantial amounts of biomedical literature, drug package inserts, clinical guidelines, and lay health information — using deep learning methods that enable them to generate statistically plausible, contextually appropriate text in response to a prompt. They excel at tasks involving language fluency and pattern recognition: summarizing a topic at a general level, explaining a pharmacological concept, generating a list of potential drug interactions, or drafting patient education materials. For tasks where a conceptually plausible but not necessarily factually accurate response is acceptable, LLMs can be highly productive tools.

However, LLMs have critical limitations that make them unsuitable as primary sources of clinical drug information without expert verification. First, hallucination: LLMs generate responses based on patterns in training data rather than explicit knowledge retrieval, and they will confidently generate plausible-sounding but factually incorrect information — including fictitious drug interactions, non-existent references, incorrect dosing, and inaccurate descriptions of drug mechanisms — without any indication that the information may be incorrect. Second, training data cutoff: LLMs have a training data cutoff date and cannot access information published after that date; for questions about recently approved drugs, updated dosing guidelines, new drug interactions, or current drug shortages, they will provide outdated or no information. Third, lack of transparency: LLMs cannot reproducibly identify their sources, making it impossible to verify the evidence base for their responses or to critically appraise the quality of the underlying evidence.

Drug Information Responses — Structure and Communication

The ability to formulate, research, and communicate a systematic drug information response is a core pharmacy skill assessed throughout the PharmD program and exercised daily in clinical practice. A structured drug information response follows a framework that mirrors the clinical reasoning process: collecting information (what is the question?), assessing information (searching and critically appraising the evidence), creating a response (synthesizing the evidence into a clear, actionable, patient-specific recommendation), and documenting and following up.

The Modified Systematic Approach to drug information questions, adapted for pharmacy practice, involves seven steps: (1) securing demographics of the requestor — who is asking and why (a prescriber making a prescribing decision? a patient trying to understand their medication? a nurse asking about administration? the context shapes the appropriate level of detail and the format of the response); (2) obtaining background information — what is the clinical context? patient-specific details about the patient for whom the question arises (age, sex, renal/hepatic function, other medications, allergies, indication for the drug); (3) determining and categorizing the ultimate question — what is really being asked? is it a dosing question? a drug interaction question? an adverse effect question? an efficacy question? (4) developing a search strategy and identifying resources — which tertiary sources and secondary databases are most likely to address this specific question type? (5) performing the search and evaluating the literature — critically appraising the quality and relevance of identified evidence; (6) formulating and communicating the response — synthesizing the evidence into a clear, concise, appropriately qualified response tailored to the requestor’s needs; and (7) following up — checking whether the response was helpful and whether additional information is needed.

Chapter 10: Epidemiologic Concepts in Drug Information Practice

Causality Assessment — The Bradford Hill Criteria

In pharmacovigilance, drug information, and clinical practice, the question of whether an observed adverse event is caused by a drug — rather than by the underlying disease, a co-medication, or chance — is one of the most practically important and intellectually challenging in drug therapy evaluation. The Bradford Hill criteria, proposed by the epidemiologist Austin Bradford Hill in a 1965 lecture, provide a framework for assessing causality in epidemiological associations that is widely applied to pharmacovigilance and adverse drug reaction assessment.

The Bradford Hill criteria include: strength of association (stronger associations are more likely to be causal — an OR of 10 is more convincingly causal than an OR of 1.2, which could easily be due to confounding); consistency (has the association been observed in multiple studies by different researchers in different settings?); specificity (is the association specific to a particular exposure-outcome pair?); temporality (does the exposure precede the outcome in time? — this is the only necessary criterion, as an effect cannot precede its cause); biological gradient (is there a dose-response relationship — do higher exposures produce greater effects?); plausibility (is there a biologically plausible mechanism?); coherence (does the association cohere with the natural history and biology of the disease?); experimental evidence (do experimental studies — particularly RCTs — support the association?); and analogy (have similar exposures produced similar effects?). None of these criteria is individually sufficient for establishing causality, and their application requires judgment rather than mechanical scoring.

The Naranjo Adverse Drug Reaction Probability Scale

In clinical practice, pharmacists frequently encounter the question of whether a patient’s new symptom, laboratory abnormality, or clinical event is an adverse drug reaction (ADR). The Naranjo Adverse Drug Reaction Probability Scale is a simple, validated instrument that assigns a numerical score to an adverse drug event based on a series of questions about the event, producing a categorical classification of the probability that the event is drug-related.

The Naranjo scale consists of 10 questions: (1) Are there previous conclusive reports on this reaction? (+1 yes, 0 do not know, 0 no); (2) Did the adverse event appear after the suspected drug was given? (+2 yes, -1 no, 0 do not know); (3) Did the adverse reaction improve when the drug was discontinued or a specific antagonist was given? (+1 yes, 0 do not know, -1 no); (4) Did the adverse reaction reappear when the drug was re-administered? (+2 yes, -1 no, 0 do not know, 0 not applicable); (5) Are there alternative causes that could on their own have caused the reaction? (-1 yes, +2 no, 0 do not know); (6) Did the reaction reappear when a placebo was given? (-1 yes, +1 no, 0 do not know); (7) Was the drug detected in the blood (or other fluids) in concentrations known to be toxic? (+1 yes, 0 do not know, 0 no); (8) Was the reaction more severe when the dose was increased or less severe when the dose was decreased? (+1 yes, 0 do not know, 0 not applicable); (9) Did the patient have a similar reaction to the same or similar drugs in any previous exposure? (+1 yes, 0 do not know, 0 no); (10) Was the adverse event confirmed by any objective evidence? (+1 yes, 0 no). Total scores of 9 or above indicate definite ADR; 5 to 8 probable ADR; 1 to 4 possible ADR; 0 or below doubtful ADR.

The Naranjo scale is useful as a standardized tool for documenting ADR probability assessments in medical records and research — it produces a reproducible score that facilitates communication among clinicians and supports pharmacovigilance reporting. It has limitations: it was developed for use with drug-induced disease outcomes rather than all adverse events; it does not account for immunological mechanisms of delayed hypersensitivity reactions; and its scores should be interpreted as a guide to judgment rather than a replacement for it. The World Health Organization-Uppsala Monitoring Centre (WHO-UMC) causality classification system uses a similar categorical approach (certain, probable, possible, unlikely, conditional/unclassified, unassessable/unclassifiable) and is the standard used in international pharmacovigilance reporting.

Epidemiologic Measures of Drug Use — Defined Daily Dose

The defined daily dose (DDD) is a technical unit developed by the WHO for drug utilization studies. The DDD is defined as the assumed average maintenance dose per day for a drug used for its main indication in adults. It is a measurement unit designed to allow comparison of drug utilization across different drugs, countries, healthcare systems, and time periods — normalizing for the fact that drugs within the same therapeutic class may have very different dosage regimens. DDDs are expressed as number of DDDs per 1,000 inhabitant-days, which allows estimation of the proportion of a population that receives daily treatment with a given drug.

For example, if a defined population shows 20 DDD/1,000 inhabitants/day for metformin, this means that on average, 2% of the population (20 out of 1,000) receives the DDD of metformin daily. DDDs are assigned by the WHO Collaborating Centre for Drug Statistics Methodology and are published in the ATC/DDD Index; they are not intended to reflect recommended or prescribed doses for individual patients, and they do not account for dosing adjustments for renal impairment or other patient-specific factors. Nonetheless, DDD-based analyses are widely used in pharmacoepidemiology to describe population-level drug use patterns, track trends in prescribing over time, compare drug use between jurisdictions, and monitor the implementation of prescribing guidelines.

Chapter 11: Writing and Communicating Drug Information

Drug Information Response Writing

A high-quality written drug information response synthesizes the best available evidence into a clear, actionable, well-referenced answer to a specific question, calibrated to the needs of the requestor and the clinical context. Writing such responses is a skill that requires practice, critical self-evaluation, and feedback — and it is assessed throughout the PHARM 155 course through the Drug Information Assignment.

The structure of a formal drug information response typically includes: an identification of the requestor and the clinical question as understood; a brief description of the search strategy used to identify relevant evidence; a critical summary of the relevant evidence, organized logically (typically from highest to lowest quality evidence, or from most to least relevant evidence); a clear, direct answer to the question with explicit acknowledgement of uncertainty; and a list of references in a recognized citation format. For clinical drug information questions, the response should always include a statement about how the evidence applies to the specific patient, including patient-specific factors that may modify the general recommendation (renal function, drug interactions, allergy history, patient preference).

A common pitfall in drug information responses is failing to answer the question directly. After a thorough literature review and detailed evidence summary, some responses become so preoccupied with methodological nuances that they never clearly state what the pharmacist actually recommends. The requestor — whether a physician making a prescribing decision or a patient trying to understand their medication — needs a clear, actionable answer, appropriately qualified by uncertainty. “Based on the available evidence, I recommend X, with Y monitoring for Z adverse effect; if X is not available or appropriate, Y is an acceptable alternative” is more useful than “the evidence is mixed and more studies are needed.” Evidence-based practice requires making decisions under uncertainty, not deferring all decisions until perfect evidence is available.

Communicating Uncertainty and Evidence Quality

A recurring challenge in drug information practice is communicating uncertainty accurately without either overstating confidence (which can lead to harm if the information is wrong) or being so hedged that the response is clinically useless. The GRADE system’s distinction between “strong recommendations” and “conditional recommendations” provides a useful vocabulary: a strong recommendation implies that the benefits clearly outweigh the harms for most patients and that the evidence is of sufficient quality that one can be confident; a conditional recommendation implies that the evidence is of lower quality, that patient values and preferences vary, or that resource implications are significant, such that different choices may be appropriate for different patients.

Specific language for communicating evidence quality levels: for high-quality evidence from consistent RCTs, language such as “evidence strongly supports” or “randomized controlled trials consistently demonstrate” is appropriate; for moderate-quality evidence from RCTs with important limitations or consistent observational studies, “evidence suggests” or “observational studies indicate”; for low-quality evidence, “limited evidence suggests” or “case reports indicate”; for very low quality (expert opinion, case reports), “experts recommend” or “based on pharmacological principles, it is reasonable to expect.” Citing specific studies with their key outcomes — rather than making unsourced assertions — gives the requestor the information needed to exercise their own critical judgment, respects their autonomy as a professional, and maintains the pharmacist’s credibility as an evidence-based practitioner.

Chapter 12: Advanced Critical Appraisal — Randomized Controlled Trials

The Architecture of a Well-Designed RCT

The randomized controlled trial is the gold standard for establishing efficacy of therapeutic interventions because randomization, when successful, distributes both known and unknown confounders equally between treatment groups, allowing a causal interpretation of the observed treatment difference. Understanding the internal architecture of a well-designed RCT — and, equally, recognizing the sources of bias when design elements are inadequately implemented — is the central competency of clinical critical appraisal.

Random allocation of participants to treatment groups is the defining feature of the RCT. True randomization uses a random number generator or a randomization table to determine treatment assignment, ensuring that each participant has an equal (or prespecified probability) of being assigned to each group, and that the assignment is not predictable in advance. Simple randomization (purely random assignment) works well in large trials but can produce imbalanced group sizes in small trials by chance; restricted randomization methods — block randomization (assigning participants in alternating blocks to ensure balanced group sizes at any given point in enrollment), stratified randomization (performing separate randomization sequences within pre-defined patient subgroups such as disease severity or age strata), and minimization (dynamically adjusting assignment to minimize imbalances in prognostic factors) — are used to improve balance. The distinction between randomization (the process of generating the allocation sequence) and allocation concealment (preventing investigators from knowing the allocation before a participant is enrolled) is critically important. Allocation concealment prevents selection bias — the tendency of investigators who know the allocation sequence to selectively enroll or exclude patients based on where they would be assigned. Methods for ensuring allocation concealment include central telephone or web-based randomization systems, sequentially numbered, sealed, opaque envelopes, and pharmacy-controlled dispensing of consecutively numbered treatment kits; studies using inadequate allocation concealment (e.g., open lists visible to investigators) have been shown to overestimate treatment effects by 30-40% compared to studies with adequate concealment.

Blinding (masking) prevents knowledge of treatment assignment from influencing outcome assessment, participant behavior, or clinician management decisions. A single-blind trial blinds participants but not investigators; a double-blind trial blinds both participants and outcome assessors (and ideally all study personnel); a triple-blind trial additionally blinds the data safety monitoring committee. Double-blinding is achievable when the experimental and control interventions can be made to appear identical — identical capsule appearance, taste, injection volume, and administration schedule. For surgical or behavioral interventions, blinding is often impossible; in such cases, using objective (rather than patient-reported) primary endpoints and blinded outcome assessors reduces performance and detection bias.

The CONSORT (Consolidated Standards of Reporting Trials) statement provides a 25-item checklist and flow diagram that specify what information must be reported in a published RCT to allow readers to assess the risk of bias and interpret the results. The CONSORT flow diagram shows the number of participants screened, enrolled, randomized, lost to follow-up, and analyzed at each stage — allowing assessment of whether losses to follow-up are substantial enough to threaten the validity of the ITT analysis (a commonly used rule of thumb is that differential loss to follow-up exceeding 20% seriously threatens validity). The risk of bias assessment tools — including the Cochrane RoB 2 tool (for RCTs) and the ROBINS-I tool (for non-randomized studies) — provide structured frameworks for assessing bias in the domains of randomization, allocation concealment, blinding, incomplete outcome data, selective outcome reporting, and other potential sources of bias, and are used in systematic reviews to weight and interpret the contributing studies.

Sample Size and Statistical Power

A critical aspect of RCT design — and a common source of misleading negative results — is the sample size calculation. A trial that is too small has insufficient statistical power to detect a true treatment difference of the size that matters clinically; such underpowered trials produce a negative result not because the treatment is ineffective, but because the study could not reliably detect a real effect. Conversely, a trial that is very large may detect statistically significant treatment differences that are so small as to be clinically meaningless — a reminder that statistical significance and clinical significance are distinct concepts.

Sample size calculation requires specification of four parameters: the expected event rate or mean value in the control group (obtained from prior studies or registry data); the minimum clinically important difference (MCID) in the primary outcome — the smallest difference that would change clinical practice; the desired level of statistical significance, conventionally alpha = 0.05 (the acceptable probability of a Type I error — falsely concluding that a treatment works when it does not); and the desired power, conventionally 80% or 90% (the probability of detecting a true difference of the specified magnitude if it exists — 1 minus the probability of a Type II error, or falsely concluding that a treatment does not work when it does). Sample size and power calculations assume that the primary outcome is the one pre-specified as primary; trials that change their primary endpoint after data collection are engaging in “outcome switching” — a form of selective reporting bias. The ORBIT (Outcome Reporting Bias in Trials) studies have documented that outcome switching is common in published trials and consistently biases toward positive results by selecting for analysis whichever outcomes showed favorable results.

Superiority, Non-Inferiority, and Equivalence Trials

The vast majority of early-phase drug development trials are superiority trials — designed to demonstrate that the new drug produces a better outcome than the comparator (placebo or active control). However, once a drug is established as effective, many subsequent trials test whether a new agent is non-inferior to the established agent — not better, but no worse by more than a pre-specified margin, the non-inferiority margin (delta). Non-inferiority trials are conducted when the new agent offers potential advantages in safety, tolerability, cost, route of administration, or patient convenience that would justify its use even if it were not more effective.

The non-inferiority margin is the most critical and contested element of a non-inferiority trial design. Delta must be selected based on the clinical judgment of what difference in efficacy would be acceptable given the potential advantages of the new agent — a judgment that is inherently subjective and that different stakeholders (regulators, clinicians, patients, payers) may view differently. If delta is set too large (accepting a substantial inferiority), the trial may demonstrate “non-inferiority” for a drug that is meaningfully less effective; if set too small, the trial may be underpowered. Regulatory guidance from both Health Canada and the FDA specifies that the non-inferiority margin must not exceed the entire estimated effect of the active control versus placebo in historical trials — a requirement called preservation of effect — to ensure that a chain of non-inferiority trials does not progressively dilute therapeutic standards.

Chapter 13: Observational Study Designs — Strengths, Limitations, and Interpretation

Cohort Studies — Prospective and Retrospective Designs

Cohort studies follow groups of individuals defined by their exposure status — exposed to a drug or other factor versus unexposed — and compare the incidence of outcomes over time. In a prospective cohort study, the cohort is assembled before outcomes have occurred, and participants are followed forward in time; in a retrospective cohort study, both exposure and outcome data are obtained from historical records, but the temporal sequence is the same (exposure precedes outcome). Cohort studies are ideally suited for studying the etiology of disease and the long-term effects of drug exposures — particularly when RCT is not feasible for ethical reasons (e.g., studying the effects of smoking, alcohol use, or medication non-adherence), is too slow (studying 20-year cardiovascular outcomes), or is too expensive.

The major advantage of prospective cohort studies is their ability to assess multiple outcomes for a single exposure, to directly measure incidence rates, and to collect detailed exposure information at baseline (reducing recall bias compared to retrospective designs). The major limitation is confounding — the systematic difference in baseline characteristics between exposed and unexposed individuals that can create spurious associations. In pharmacoepidemiological cohort studies — which commonly use insurance claims or electronic health record databases rather than prospectively collected data — confounding by indication is a particular concern: the reasons that a patient receives a particular drug (i.e., the indication) are themselves predictors of outcomes. For example, patients prescribed statins have higher cardiovascular risk than patients not prescribed statins; a naive comparison of cardiovascular events in statin users versus non-users would show higher event rates in statin users — not because statins cause cardiovascular disease, but because they are prescribed to patients at higher cardiovascular risk. Methods for controlling confounding in observational studies include multivariate regression (adjusting for measured confounders), propensity score methods (matching or stratifying on the estimated probability of being exposed, given measured baseline characteristics), instrumental variable analysis (using a variable that predicts exposure but is unrelated to outcome through any path other than exposure — such as geographic variation in prescribing rates as an instrument for studying medication effects), and active comparator designs (comparing the drug of interest to another drug used for the same indication, reducing confounding by indication).

Case-Control Studies — Efficient Study of Rare Outcomes

Case-control studies begin with the selection of cases — individuals who have experienced the outcome of interest — and controls — individuals from the same base population who have not experienced the outcome — and compare the frequency of past exposures between the two groups. This retrospective design is particularly efficient for studying rare outcomes (such as rare drug-induced adverse events) because the case-control design allows studying a large number of outcomes relative to the sample size, without following an entire cohort forward in time until a sufficient number of rare events accumulate.

The critical challenge in case-control study design is the selection of appropriate controls — they must be sampled from the same population that gave rise to the cases, without selection based on the exposure of interest. Hospital-based controls — selecting controls from among patients hospitalized for conditions other than the outcome of interest — are convenient but can introduce Berkson’s bias (a form of selection bias in which hospitalized controls have different exposure patterns than the community population from which cases arise). Population-based controls — randomly sampled from the community — are methodologically superior but more expensive to recruit. For drug-related case-control studies using administrative data, incident density sampling (selecting controls matched to cases on calendar time, age, sex, and geographic region, and allowing the same person to serve as a control at one time and later be selected as a case) most closely approximates the ideal design.

The measure of association produced by a case-control study is the odds ratio (OR) — the ratio of the odds of exposure among cases to the odds of exposure among controls:

where a is the number of exposed cases, b is the number of exposed controls, c is the number of unexposed cases, and d is the number of unexposed controls. When the outcome is rare (less than 10%), the OR closely approximates the relative risk (RR); for common outcomes, the OR overestimates the RR. Case-control studies cannot directly measure incidence rates or risk differences; they can only estimate relative measures of association. Recall bias — the differential ability of cases (who have experienced a disease) to recall past exposures compared to controls — is a potential source of information bias in case-control studies using self-reported exposure data; it is minimized in studies using objective exposure records (pharmacy dispensing records, hospital medication administration records).

Chapter 14: Drug Information in Specialized Clinical Contexts

Therapeutic Drug Monitoring — Pharmacokinetic Drug Information

Therapeutic drug monitoring (TDM) is the practice of measuring drug concentrations in a patient’s blood or other biological fluid and using the result to individualize drug dosing — adjusting the dose to maintain concentrations within a defined therapeutic range that optimizes efficacy and minimizes toxicity. TDM is clinically indicated for drugs with narrow therapeutic indices (where small differences in concentration produce large differences in effect), for drugs with unpredictable pharmacokinetics due to interindividual variability, and for situations where patient-specific factors (organ impairment, drug interactions, genetic polymorphisms) significantly alter pharmacokinetics.

The generation of a TDM-guided dosing recommendation requires integration of the measured concentration with knowledge of the pharmacokinetic model applicable to the drug, the expected concentration-time profile from the prescribed regimen, and the patient-specific factors that may alter clearance or volume of distribution. For vancomycin, the current AUC-guided approach (recommended by the 2020 ASHP/IDSA/SIDP vancomycin therapeutic monitoring consensus guidelines, which have been adopted in most Canadian hospitals) calculates the 24-hour area under the concentration-time curve (AUC₂₄) using two timed serum concentration measurements (a peak drawn 1-2 hours after completion of the infusion and a trough drawn 1 hour before the next dose) and Bayesian analysis, with a target AUC₂₄ of 400-600 mg·h/L (for serious MRSA infections). This approach replaces the older trough-only monitoring strategy (target trough 15-20 mg/L for serious infections), which was associated with nephrotoxicity — likely because maintaining troughs of 15-20 mg/L requires pushing the peak into the nephrotoxic range in many patients.

The Bayesian approach to TDM — now implemented in commercial software platforms such as DoseMe and InsightRx — uses population pharmacokinetic models to generate a prior probability distribution for the patient’s pharmacokinetic parameters (clearance, volume of distribution) based on patient-specific covariates (age, sex, weight, renal function), then updates this prior using the individual patient’s measured concentration(s) to generate a posterior estimate of the patient’s individual pharmacokinetic parameters and a predicted dosing regimen to achieve the target exposure. Bayesian individualization is superior to the traditional nomogram approach (which uses a fixed dose adjustment factor based on creatinine clearance) because it explicitly accounts for the uncertainty in parameter estimates and updates the prediction as more concentration data become available. Pharmacy-led TDM services — in which clinical pharmacists perform concentration interpretation, dosing calculations, and recommendations — are a well-established model of pharmacist clinical practice in Canadian hospital pharmacy.

Drug Information in Oncology — Special Challenges

Oncology pharmacy presents unique drug information challenges because the evidence base for cancer treatments is structured differently from other therapeutic areas, the stakes of suboptimal drug information are exceptionally high (both under-treatment and over-treatment can cause severe harm), and the rapidly evolving nature of the field means that information can become outdated quickly. Clinical pharmacists practicing in oncology settings must be familiar with oncology-specific drug information resources, the interpretation of oncology clinical trial data, and the clinical and regulatory pathways for new oncology drugs.

Oncology clinical trials increasingly use surrogate endpoints — endpoints that are measurable earlier than overall survival (OS) but are expected to predict OS — as primary endpoints to accelerate drug development. Common surrogate endpoints in oncology include: progression-free survival (PFS) — the time from randomization to disease progression (defined by imaging criteria such as RECIST) or death; objective response rate (ORR) — the proportion of patients with measurable tumor shrinkage meeting a pre-specified size threshold; and disease-free survival (DFS) — the time from definitive treatment to disease recurrence or death, used in the adjuvant setting. The regulatory acceptance of surrogate endpoints for approval — and particularly for accelerated approval (FDA) or NOC/c (Health Canada) — has been extensively debated, because surrogates do not always reliably predict OS. PFS improvement predicts OS improvement reliably in some cancer settings (e.g., colon cancer) but poorly in others (e.g., some breast cancer settings); drug information practitioners must understand which surrogate endpoints are validated in which settings and interpret trial results accordingly.

Oncology drug information must also address the complex practical aspects of cancer drug use: the preparation and stability of antineoplastic agents (many of which have limited stability in solution and require specialized preparation conditions in negative-pressure isolators to protect the compounder from exposure); dose calculation based on body surface area (BSA) or weight-based dosing; dose capping for obese patients (an area of ongoing controversy — historical practice of capping BSA-based doses in obese patients is not supported by pharmacokinetic evidence and may result in underdosing of cancer drugs, with negative impact on outcomes); management of drug interactions with oncology agents (the CYP and P-gp interaction profiles of targeted agents and many chemotherapy drugs are complex and clinically significant); and the management of oncology emergencies such as tumor lysis syndrome (TLS) — the metabolic emergency caused by massive cancer cell death releasing intracellular contents — for which allopurinol and rasburicase prophylaxis, hyperhydration, and electrolyte monitoring protocols are evidence-based standards.

Drug Information Resources for the Canadian Pharmacist

The landscape of drug information resources available to Canadian pharmacists includes regulatory resources unique to the Canadian context alongside international resources of general applicability. The Compendium of Pharmaceuticals and Specialties (CPS, now RxTx) — published by the Canadian Pharmacists Association — is the Canadian equivalent of the US Physicians’ Desk Reference (PDR) or the British National Formulary (BNF), compiling the official product monographs of Health Canada–approved drugs along with clinical monographs, patient information leaflets, and clinical decision-support tools. Access to RxTx is included in CPhA membership and is available through most pharmacy practice sites in Canada; it is the first-line reference for Canadian-specific drug information including approved indications, dosing, drug interactions, and Canadian regulatory information.

The Drug Product Database (DPD) maintained by Health Canada is the official Canadian source for information about all drugs approved for sale in Canada — including product monographs, market status (marketed or discontinued), drug identification numbers (DINs), and active ingredient information. The DPD is searchable online and through API for integration into digital clinical tools; it is the definitive authority for questions about whether a drug is currently approved in Canada, under what brand names, and for what formulations and strengths. For questions about the regulatory status of drugs — including drugs approved in other countries but not in Canada, drugs approved under special access provisions (Health Canada’s Special Access Program, which allows physicians to request access to unapproved drugs for patients with serious or life-threatening conditions where no comparable treatment is available), and drugs approved with conditions — the DPD and the Health Canada drug product listings are the authoritative resources.

Chapter 15: Health Technology Assessment and Drug Formulary Management

The Canadian Drug Review Process — CADTH and INESSS

Health technology assessment (HTA) is the systematic evaluation of the properties, effects, and impacts of health technologies — including drugs, medical devices, diagnostics, and health interventions — to inform coverage and reimbursement decisions by public payers. In Canada, the primary federal HTA body for drugs is the Canadian Drug Review function of CADTH (the Canadian Drug and Devices Agency, formerly the Canadian Agency for Drugs and Technologies in Health). CADTH’s Drug Reimbursement Review (DRR) evaluates new drugs and new indications for existing drugs and produces reimbursement recommendations that inform provincial and territorial public drug benefit plans.

The CADTH reimbursement review process begins when a drug manufacturer submits an application for listing on the public drug formulary, accompanied by a comprehensive clinical evidence package and a pharmacoeconomic submission. CADTH’s review assesses clinical benefit (does the drug produce meaningful health benefit compared to the best current alternative?), harms (what are the adverse effects and their clinical significance?), and cost-effectiveness (is the drug’s health benefit worth its cost compared to available alternatives?). The economic analysis calculates an incremental cost-effectiveness ratio (ICER) — the additional cost per additional unit of health outcome (typically measured in quality-adjusted life years [QALYs]) that the new drug produces compared to the comparator. CADTH does not use a fixed cost-effectiveness threshold; instead, it issues a recommendation (reimburse, reimburse with conditions, do not reimburse) based on a holistic assessment that considers clinical evidence quality, ICER estimates, uncertainty, patient perspectives, and equity considerations.

Pharmacoeconomic Methods — Cost-Effectiveness and Cost-Utility Analysis

Pharmacoeconomics is the discipline that applies economic principles and methods to the valuation of drug therapies — assessing the relationship between costs and outcomes (health consequences) to inform decisions about which treatments represent good value for money. The major pharmacoeconomic study types differ in how they measure and value outcomes.

Cost-minimization analysis (CMA) compares the costs of two treatments that are assumed to be equivalent in effectiveness; it is appropriate only when prior evidence (typically from a non-inferiority or equivalence trial) has established that the two treatments produce identical outcomes. CMA is straightforward but requires strong evidence of equivalence that is rarely available; it is increasingly being replaced by cost-effectiveness analysis even in situations where drugs are expected to be similar in efficacy, because truly identical outcomes are uncommon and small differences in effectiveness matter at the population level.

Cost-effectiveness analysis (CEA) measures outcomes in natural units (life-years gained, cases of disease prevented, hospitalizations avoided, response achieved) and calculates an incremental cost-effectiveness ratio (ICER):

The result — cost per life-year gained, cost per response, etc. — is interpreted relative to a threshold representing the decision-maker’s willingness to pay for a given outcome. CEA is useful for comparing drugs within a single therapeutic area but cannot directly compare across different disease areas with different natural outcome units.

Cost-utility analysis (CUA) measures outcomes in quality-adjusted life years (QALYs) — a unit that integrates the quantity and quality of life produced by a treatment. One QALY represents one year of life in perfect health; a year of life in a state with quality 0.5 (halfway between death and perfect health) contributes 0.5 QALYs. Health state utilities — the quality weights used to calculate QALYs — are typically measured using preference-based instruments (the EQ-5D, SF-6D, HUI) administered to patients or valued using time trade-off or standard gamble methods in population surveys. CUA allows comparison of treatments across different disease areas on a common QALY scale; the ICER (cost per QALY gained) is then compared against an implicit or explicit willingness-to-pay threshold. CADTH’s pharmacoeconomic guidelines recommend CUA as the preferred form of economic analysis for drug submissions; they note, without specifying a fixed threshold, that an ICER below approximately $50,000 per QALY is generally considered cost-effective in the Canadian public healthcare context, while ICERs above $100,000-$150,000 per QALY are considered unlikely to represent good value, and drugs with ICERs in the intermediate zone require judgment based on clinical context and uncertainty.

The QALY framework, while dominant in HTA methodology, is not without critics. Concerns include: that QALYs measured from population surveys may not reflect the values of patients living with specific conditions; that the QALY framework systematically undervalues treatments for rare diseases, where the small patient population makes large QALY gains per patient economically insufficient to meet conventional thresholds; that end-of-life treatments may warrant a premium above standard cost-effectiveness thresholds because of the special ethical significance of life-extending treatments for terminally ill patients; and that the assumption of consistent utility values across individuals ignores heterogeneity in patient values and preferences. Alternative multi-criteria decision analysis (MCDA) frameworks — which explicitly incorporate multiple dimensions of value (clinical benefit, severity of disease, unmet need, innovation, patient and caregiver burden, equity) — are being developed as complements or alternatives to QALY-based CUA in HTA decision-making.

Budget Impact Analysis and Real-World Evidence

Budget impact analysis (BIA) addresses a question distinct from cost-effectiveness analysis: not “is this drug worth its price?” but “what will it cost the payer to list this drug, given the expected number of patients, their treatment patterns, and the drug’s price?” BIA is a prospective financial planning exercise — estimating the total budget impact of adopting a new drug into a formulary over a defined time horizon (typically 1, 3, and 5 years) — and is a mandatory component of drug formulary submissions in Canada and most developed countries. A drug with a favorable ICER (cost-effective at the individual level) may nevertheless have an unacceptably large budget impact if the patient population is very large, the drug very expensive, or the treatment duration very long; payers must balance cost-effectiveness of individual treatments with the aggregate affordability of the portfolio of treatments they fund.

Real-world evidence (RWE) — evidence generated from analysis of real-world data (RWD) sources such as electronic health records, administrative claims databases, patient registries, and wearable device data — is increasingly important in HTA and formulary management as a complement to clinical trial evidence. RCT evidence has high internal validity but limited external validity — it tells us what the drug does under ideal trial conditions, not necessarily what it will do in routine clinical practice in diverse populations. RWE provides evidence on effectiveness (how the drug performs in real-world populations including those excluded from trials), safety (detecting adverse effects too rare for trial detection), comparative effectiveness (comparing drugs that have never been compared head-to-head in a trial), and utilization patterns (how drugs are actually prescribed, for what indications, at what doses, for how long). CADTH’s Real-World Evidence report series and the Canadian Network for Observational Drug Effect Studies (CNODES) — a network of eight provincial and one federal administrative data nodes representing over 40 million Canadians — are major Canadian real-world evidence infrastructure resources that contribute to formulary and coverage decisions.