Worked Design Of A Genotype-Guided Prescribing Trial

From a Prescribing Question to a Trial

A pharmacogenomic result can influence which drug is prescribed, its starting dose, or how quickly a dose is adjusted. To learn whether using that information improves outcomes, the trial must compare two well-defined prescribing strategies, not simply compare people with different genotypes. Random assignment can then separate the effect of offering the genotype-guided strategy from differences in the patients who happen to carry particular genotypes.

The design below considers a hypothetical trial for adults beginning antidepressant treatment. A research genotype result identifies one of two prescribing categories, and a prespecified guideline uses that result to select or dose treatment. The comparison is with usual care. The names, data, and numerical results in the examples are hypothetical.

Specify the Strategies Before Randomization

“Use pharmacogenomics” is not a sufficiently precise intervention description. The protocol needs to state which variants are tested, how results are converted into prescribing categories, and what action follows each category. For the hypothetical trial, suppose the laboratory classifies results as category A or category B. In the genotype-guided group, the prescriber receives the category and follows a written table specifying preferred first-line drugs and starting-dose ranges for each category. The table also states how to proceed if the preferred option is unsuitable.

In the usual-care group, prescribers select and adjust treatment using the information ordinarily available to them, but do not receive the research genotype report. The protocol should describe any safety-related exception, such as a rule for disclosing a result if a prespecified urgent finding is identified. That rule must apply consistently and be recorded. Both groups need the same access to routine follow-up and clinical support; otherwise, the trial would compare a bundle of differences rather than principally the prescribing strategies.

The treatment is assignment to a strategy, not perfect compliance with every recommendation. A prescriber may depart from the guideline for a clinical reason, and a patient may stop or switch medication. Those events are important to record, but they do not change the randomized assignment used in the primary comparison.

Choose an Endpoint That Works in Both Groups

The endpoint should measure a clinically interpretable outcome at a specified time, using the same procedure in each arm. Here, suppose the primary endpoint is change in depression symptom score from baseline to week 12, measured with a prespecified validated scale on which higher scores indicate more severe symptoms. The score is collected even if a participant stops or switches the initially prescribed medication. Thus the outcome reflects the consequences of assignment to the strategy, including treatment changes that occur in practice.

For participant \(i\), define \(Y_i\) as the week-12 score minus the baseline score, in scale points. A negative change indicates symptom improvement. The protocol should also specify the assessment window, who collects the score, and how a score outside the window is handled. These details reduce the risk that the endpoint is measured differently because a participant was assigned to one strategy rather than the other.

The primary endpoint is not the number of genotype recommendations followed. That is a process measure: it can help describe how the strategy was delivered, but it does not by itself establish whether patients’ symptoms improved. Prespecified secondary outcomes may include adverse effects or treatment changes, but adding outcomes does not change what the primary contrast is intended to answer.

Define the Population and Primary Contrast

Let \(A_i=1\) denote assignment to genotype-guided prescribing and \(A_i=0\) denote assignment to usual care. Let \(G_i\) denote the participant’s prespecified genotype category, and let \(Y_i(1)\) and \(Y_i(0)\) denote the week-12 change that participant would have under assignment to each strategy. Only one of these potential outcomes is observed for a participant. The primary analysis population is all eligible participants who are randomized, analyzed according to their assigned group, regardless of treatment received afterward.

Suppose randomization is stratified by genotype category. The overall primary contrast is the mean change under guided assignment minus the mean change under usual-care assignment, averaged over the genotype-category distribution of the randomized population. If \(w_g\) is the proportion of randomized participants in category \(g\), and \(\mu_{ag}=E[Y(a)\mid G=g]\), then:

The contrast is measured in symptom-score points. Because lower scores indicate improvement, a negative value favors assignment to the genotype-guided strategy. The weights define the population to which this overall contrast applies; using the randomized population’s genotype distribution makes the target explicit.

With random assignment within genotype strata, assignment is independent of the potential outcomes within each stratum. Under consistency (the assigned strategy corresponds to the strategy defined in the protocol) and no interference between participants, the observed mean outcome in each assigned group and genotype stratum estimates the corresponding potential-outcome mean. Weighting the within-stratum contrasts by \(w_g\) therefore estimates the stated overall contrast. This is the key justification for the primary comparison; it does not require participants to follow their assigned strategy perfectly.

A trial may improve precision by adjusting for baseline score and randomization strata, such as genotype category and site. If it does, the analysis should still report a contrast targeting the overall randomized population, rather than silently changing the target to a different group. Week-12 scores should be sought after treatment discontinuation or switching. If an endpoint remains missing, the protocol needs a prespecified approach, such as multiple imputation under an explicit missing-at-random assumption, together with sensitivity analyses that examine departures from that assumption. Excluding everyone with a missing score without considering why it is missing can undermine the randomized comparison.

Estimate the Stratified Contrast

Plan Enrollment Around a Prespecified Target

A sample-size calculation requires a target difference worth detecting for the design, an assumed outcome variability, a significance level, and desired power. For a simple planning approximation with equal allocation, a continuous endpoint, independent participants, a two-sided test, and similar variance in both groups, the required number per group is:

Here \(\sigma\) is the assumed standard deviation of score change, \(\delta\) is the target mean difference, \(\alpha\) is the type I error rate, and \(1-\beta\) is power. This approximation is for planning; the final calculation should reflect the intended stratified analysis, any important design features, and the planned handling of missing outcomes.

Why the Strategy-Level Contrast Matters

The central design choice is to compare assignment to two reproducible care strategies in a defined population. A genotype-stratified randomization scheme can ensure that both strategies are evaluated across genotype categories, while the weighted overall contrast answers the population-level question specified in advance. Category-specific differences may also be informative, but they do not replace the primary contrast and should not be treated as definitive merely because one category’s estimate is larger.

The result estimates the effect of assignment to the genotype-guided policy as delivered in the trial, including departures from the prescribing table and later treatment changes. It does not, by itself, estimate the effect of perfect adherence, prove that genotype caused a symptom difference, or establish that the strategy should be adopted in every setting.

Common Mistakes

• Comparing genotype groups instead of randomized strategies. A difference between categories A and B can reflect their differing characteristics. The primary comparison is guided assignment versus usual-care assignment, with genotype handled as specified in the design.
• Defining the intervention only as “testing.” A test result can change care only through a decision rule. Specify how results reach prescribers and what actions the strategy recommends.
• Analyzing only participants who followed the guideline. Adherence occurs after randomization and may be related to prognosis. The primary analysis retains participants in their assigned groups.
• Using an outcome available only while a participant follows the initial prescription. This can make the outcome undefined selectively. In this example, the symptom score is collected at week 12 even after treatment changes.
• Interpreting a negative change as worsening. The endpoint is week-12 minus baseline score, and higher scores mean more severe symptoms. A negative change indicates improvement; a negative guided-minus-usual contrast favors guided assignment.