Can ad hoc analyses of clinical trials help personalize treatment decisions?

The value of ad hoc analyses of clinical trials may not be clear. Ad hoc analyses are carried out after developing analysis plans and are considered exploratory in nature. Such analyses should be conducted on a ‘rational’ biological basis, rather than ‘aimless’ data dredging. Ad hoc analyses are done to generate new hypotheses that guide future research studies. Such retrospective analyses are of great value, especially if the differences for the primary outcome (adequately powered) were not detected. Analyses of positive trials can also identify populations that are treatment‐resistant or susceptible to adverse effects. Ad hoc analyses are highly underutilized to guide future studies to personalize treatment recommendations, but could have a great potential to personalize treatment decisions.

Ad‐hoc analyses of clinical trials investigating cetuximab use in colorectal cancer resulted in tailoring treatment decisions in this population. Cetuximab received US Food and Drug Administration (FDA) approval for advanced colorectal cancer treatment in February 2004 under ‘the accelerated approval of biological products regulations, 21 CFR 601.40‐46’ 1. This approval was granted based on surrogate endpoint results until adequate evidence supporting the clinical benefit of cetuximab is developed. Although a significant improvement in overall survival (OS; hazard ratio [HR] 0.77; 95% confidence interval [CI] 0.64–0.92; P = 0.005) was reported among cetuximab users 2, resistance was commonly noticed 3. These investigators relied on the lung cancer literature to understand this phenomenon, where the correlation between KRAS mutation status and response to tyrosine kinase inhibitors, other therapies directed against anti‐EGFR, has been reported 4, 5. The ad hoc analysis of cetuximab clinical trial samples showed a similar pattern of response to anti‐EGFR targeted therapy, with preferential benefit to cetuximab observed in wild‐type KRAS subgroup 3. Patients with wild‐type KRAS tumours had significant improvement in OS (HR 0.55; 95% CI 0.41–0.74; P < 0.001), compared to no significant difference (HR 0.98; P = 0.89) reported among patients with mutated KRAS tumours. Results of this and subsequent relevant studies changed colorectal cancer clinical guidelines and practice. Testing for RAS mutation status is now required to determine eligibility for anti‐EGFR targeted therapies in patients with colorectal cancer diagnosis, and even for reimbursement purposes 6.

Another example where the ‘ad hoc analysis’ approach may be useful to personalize treatment decisions is in multiple sclerosis. A recently published pivotal Phase 3 trial (ORATORIO) evaluated the safety and efficacy of ocrelizumab in primary progressive multiple sclerosis (PPMS) 7. Ocrelizumab is a humanized monoclonal antibody that selectively depletes CD20‐expressing B cells, which is thought to slow disease progression in PPMS 8. The FDA recently approved this drug for PPMS management based on results of this trial. Reviewing the study results, the primary and first secondary endpoint were a 12‐week and 24‐week confirmed disability progression, respectively. According to figures presented in the published manuscript, the curves depicting the cumulative probability of disability progression overlapped and there may not be a significance difference between ocrelizumab and placebo at these time points. Indeed, if the trial was stopped at 96 or 108 weeks, cut points used by a prior study 9, the difference between the two arms would fail to reach significance for the primary endpoint. However, significant differences reported at longer follow‐up times imply a delayed effect of ocrelizumab to show the anticipated benefits. In addition, there was underreporting of the degree of missing data, and in which variables. This is of utmost importance given the progressive nature of the disease and the method used to impute missing data (last‐observation‐carried‐forward method). Safety information is still under assessment in the open‐label phase of the study. However, the observed increase in neoplasms in the ocrelizumab group was concerning and required further evaluation.

So, how is this study relevant to our topic? The answer simply lies is the OLYMPUS trial, an earlier study that assessed the use of rituximab, another anti‐CD20, in the PPMS population 9. The OLYMPUS trial was a randomized, double‐blinded, placebo‐controlled multicentre trial study that did not show a significant difference in confirmed disease progression between rituximab and placebo through 96 weeks. However, their subgroup analyses showed a potential benefit for younger patients, specifically those having inflammatory lesions. This implies that the benefits of anti‐CD20 agents may be more evident if these agents are used in the treatment‐sensitive PPMS population for appropriate prolonged durations.

Learning from the colorectal cancer example presented in this editorial, more ad hoc analyses should be performed to identify the optimal patient–treatment combinations, in light of new relevant evidence and to guide future research. In addition, we recommend an ad hoc subgroup analysis of the ORATORIO trial to identify the targeted population that has the maximal benefit of the promising drug, ocrelizumab, with the least exposure to uncertain side effects. However, this approach is associated with obvious limitations, such as: limited power, potential biases, and possible ‘data dredging’. Ad hoc analyses can identify a target population that should be used in the design of a future adequately‐powered clinical trial.

Competing Interests

There are no competing interests to declare. Dr. Catherine Sherwin is a Senior Editor of BJCP.

Biltaji, E. , Kumar, S. S. , Enioutina, E. Y. , and Sherwin, C. M. T. (2017) Can ad hoc analyses of clinical trials help personalize treatment decisions?. Br J Clin Pharmacol, 83: 2337–2338. doi: 10.1111/bcp.13377.