Statistical analysis of clinical trial data in cancer research

<div>Abstract<p>The FDA Oncology Center of Excellence recently launched a crowdsourcing pilot to request ideas from the scientific community for research questions that FDA could address with pooled analyses of clinical trial data submitted to the agency for regulatory purposes. This effort builds on FDA’s track record of publishing pooled analyses to explore scientific questions that cannot be addressed in a single trial due to limited sample size. The research crowdsourcing pilot tested a new approach for obtaining external input on regulatory science activities, because FDA is generally unable to share patient-level data outside of the agency due to federal disclosure laws and regulations protecting different types of data submitted in regulatory applications. We received 29 submissions over the 28-day crowdsourcing campaign, including one research idea that we are exploring for possible follow up. Based on our experience with this pilot, we learned that crowdsourcing is a promising new approach to gather external input and feedback. We identified opportunities to build understanding in the external oncology community about the types of data typically included in regulatory applications and expand dissemination of published FDA pooled analyses to help inform future drug development and clinical practice.</p></div>

BackgroundA nested case-control study is an efficient design that can be embedded within an existing cohort study or randomised trial. It has a number of advantages compared to the conventional case-control design, and has the potential to answer important research questions using untapped prospectively collected data.MethodsWe demonstrate the utility of the matched nested case-control design by applying it to a secondary analysis of the Abnormal Doppler Enteral Prescription Trial. We investigated the role of milk feed type and changes in milk feed type in the development of necrotising enterocolitis in a group of 398 high risk growth-restricted preterm infants.ResultsUsing matching, we were able to generate a comparable sample of controls selected from the same population as the cases. In contrast to the standard case-control design, exposure status was ascertained prior to the outcome event occurring and the comparison between the cases and matched controls could be made at the point at which the event occurred. This enabled us to reliably investigate the temporal relationship between feed type and necrotising enterocolitis.ConclusionsA matched nested case-control study can be used to identify credible associations in a secondary analysis of clinical trial data where the exposure of interest was not randomised, and has several advantages over a standard case-control design. This method offers the potential to make reliable inferences in scenarios where it would be unethical or impractical to perform a randomised clinical trial.

Incomplete data is a prevalent complication in longitudinal studies due to individuals' drop-out before intended completion time. Currently available methods via commercial software for analyzing incomplete longitudinal data at best rely on the ignorability of the drop-outs. If the underlying missing mechanism was non-ignorable, potential bias arises in the statistical inferences. To remove the bias when the drop-out is non-ignorable, joint complete-data and drop-out models have been proposed which involve computational difficulties and untestable assumptions. Since the critical ignorability assumption is unverifiable based on the observed part of the sample, some local sensitivity indices have been proposed in the literature. Specifically, Eftekhari Mahabadi (Second-order local sensitivity to non-ignorability in Bayesian inferences. Stat Med 2018;59:55-95) proposed a second-order local sensitivity tool for Bayesian analysis of cross-sectional studies and show its better performance for handling bias compared with the first-order ones. In this paper, we aim to extend this index for the Bayesian sensitivity analysis of normal longitudinal studies with drop-outs. The index is driven based on a selection model for the drop-out mechanism and a Bayesian linear mixed-effect complete-data model. The presented formulas are calculated using the posterior estimation and draws from the simpler ignorable model. The method is illustrated via some simulation studies and sensitivity analysis of a real antidepressant clinical trial data. Overall, the numerical analysis showed that when repeated outcomes are subject to missingness, regression coefficient estimates are nearly approximated well by a linear function in the neighbourhood of MAR model, but there are a considerable amount of second-order sensitivity for the error term and random effect variances in Bayesian linear mixed-effect model framework.

The sharing of biomedical data is essential for accelerating healthcare research, fostering medical innovation, and improving patient outcomes. Such data encompasses a wide range of sensitive information, including electronic health records, genomic sequences, and clinical trial results. Despite its value, biomedical data sharing poses significant privacy risks, such as patient re-identification, unauthorized access, and regulatory non-compliance. These concerns necessitate advanced techniques that balance the need for data utility with stringent privacy protection. Machine learning (ML) has emerged as a powerful tool to facilitate privacy-preserving biomedical data sharing. This manuscript presents a comprehensive review of state-of-the-art ML-based privacy preservation methods, including differential privacy, federated learning, homomorphic encryption, secure multi-party computation, and synthetic data generation through generative models. Each technique offers unique mechanisms to protect sensitive information while enabling collaborative analysis and predictive modeling. These methods have been applied practically across various biomedical domains, including collaborative disease risk prediction and genomic research, clinical trial data analysis, remote patient monitoring, and public health surveillance. Additionally, we evaluate relevant privacy and utility metrics that assess the effectiveness of privacy guarantees and the impact on model performance. The review further examines limitations and challenges—including computational overhead, data heterogeneity, privacy-utility trade-offs, and ethical considerations—that must be addressed to ensure robust and scalable solutions. Looking forward, the manuscript highlights promising future directions, such as hybrid privacy frameworks, enhanced synthetic data generation, real-time privacy-preserving analytics, standardization of evaluation protocols, and interdisciplinary policy development. By integrating these advancements, biomedical research can achieve safer and more effective data sharing, ultimately fostering innovation while respecting patient confidentiality and trust.

Bayesian analysis of clinical trial data may provide useful information to aid in study interpretation, especially when trial evidence suggests that the benefits of an intervention are uncertain, such as in a trial that evaluated early extracorporeal membrane oxygenation (ECMO) for severe acute respiratory distress syndrome (ARDS). To demonstrate the potential utility of Bayesian analyses by estimating the posterior probability, under various assumptions, that early ECMO was associated with reduced mortality in patients with very severe ARDS in a randomized clinical trial (RCT). A post hoc Bayesian analysis of data from an RCT (ECMO to Rescue Lung Injury in Severe ARDS [EOLIA]) that included 249 patients with very severe ARDS who had been randomized to receive early ECMO (n = 124; mortality at 60 days, 35%) vs initial conventional lung-protective ventilation with the option for rescue ECMO (n = 125, mortality at 60 days, 46%). The trial was designed to detect an absolute risk reduction (ARR) of 20%, relative risk (RR) of 0.67. Statistical prior distributions were specified to represent varying levels of preexisting enthusiasm or skepticism for ECMO and by Bayesian meta-analysis of previously published studies (with downweighting to account for differences and quality between studies). The RR, credible interval (CrI), ARR, and probability of clinically important mortality benefit (varying from RR less than 1 to RR less than 0.67 and ARR from 2% or more to 20% or more) were estimated with Bayesian modeling. Combining a minimally informative prior distribution with the findings of the EOLIA trial, the posterior probability of RR less than 1 for mortality at 60 days after randomization was 96% (RR, 0.78 [95% CrI, 0.56-1.04]); the posterior probability of RR less than 0.67 was 18%, the probability of ARR of 2% or more was 92%, and the probability of ARR of 20% or more was 2%. With a moderately enthusiastic prior, equivalent to information from a trial of 264 patients with an RR of 0.78, the estimated RR was 0.78 (95% CrI, 0.63-0.96), the probability of RR less than 1 was 99%, the probability of RR less than 0.67 was 8%, the probability of ARR of 2% or more was 97%, and the probability of ARR of 20% or more was 0%. With a strongly skeptical prior, equivalent to information from a trial of 264 patients with an RR of 1.0, the estimated RR was 0.88 (95% CrI, 0.71-1.09), the probability of RR less than 1 was 88%, the probability of RR less than 0.67 was 0%, the probability of ARR of 2% or more was 78%, and the probability of ARR of 20% or more was 0%. If the prior was informed by previous studies, the estimated RR was 0.71 (95% CrI, 0.55-0.94), the probability of RR less than 1 was 99%, the probability of RR less than 0.67 was 48%, the probability of ARR of 2% or more was 98%, and the probability of ARR of 20% or more was 4%. Post hoc Bayesian analysis of data from a randomized clinical trial of early extracorporeal membrane oxygenation compared with conventional lung-protective ventilation with the option for rescue extracorporeal membrane oxygenation among patients with very severe acute respiratory distress syndrome provides information about the posterior probability of mortality benefit under a broad set of assumptions that may help inform interpretation of the study findings.

Background: It is sometimes suggested that newly-approved cancer treatments have only marginal effectiveness, which raises questions concerning their cost-vs-benefit ratio. Such concerns appear at odds with the lower cancer-related hospitalization rate and improved survival. However, such cost-effectiveness analyses rely on population-based averages obtained from the analysis of clinical trial data. By failing to analyze data from longitudinal datasets, such assessments are unable to account for real-world patient conditions and treatment patterns in evaluating clinical and cost-effectiveness. Longitudinally surveyed clinical data has the potential to objectively reveal any association between patient outcomes and new cancer treatment utilization. Methods: We investigated the effect of being prescribed a higher proportion of new oncology drugs on quality of life, medical services use, and productivity measures as reported by the Medical Expenditure Panel Survey (MEPS, 1996–2015). General linear models with Taylor series variance estimation were applied. New oncology drugs were defined as cancer treatments marketed after the year 2000. Included subjects (N=16,677) had a solid or hematologic malignancy diagnosis (CCCodex 11–47) and available prescription data. Individual age and employment status were accounted for as covariates. All analyses were performed using SAS version 9.4 (Cary, NC). Results: Unadjusted regression data show that individuals using newer oncology treatments missed on average 2.5 (±0.3 SE) fewer days of work or school per year as compared to patients using older drugs (43% improved productivity, P<.0001). The effect persisted even after adjusting for the magnitude of the effect (P<.0001). Accounting for age, the use of newer drugs was, on average, associated with ∼35% fewer missed work or school days. Cancer patients using newer treatments had 0.06 (±0.01 SE) fewer hospital admissions/year compared to patients using older treatments (P<.0001) and spent less time in the emergency room (P<.0411) with ∼45% fewer hospitalizations. Patients using newer medicines also had fewer health-related visits (P<.0001). Conclusion: Analysis of longitudinal real-world evidence gives a more comprehensive and reliable view of the clinical and economic impact of new oncology treatments. Our data suggests significant reductions in lost work/school days, hospitalizations, and use of medical services in general.

Probabilities of hospital mortality provide meaningful information in many contexts, such as in discussions of patient prognosis by intensive care physicians, in patient stratification for analysis of clinical trial data by researchers, and in hospital reimbursement analysis by insurers. Use of probabilities as binary predictors based on a cut point can be misleading for making treatment decisions for individual patients, however, even when model performance is good overall. Alternative models for estimating severity of illness in intensive care unit (ICU) patients, while demonstrating good agreement for describing patients in the aggregate, are shown to differ considerably for individual patients. This suggests that identifying patients unlikely to benefit from ICU care by using models must be approached with considerable caution.

For a variety of reasons including poorly designed case report forms (CRFs), incomplete or invalid CRF data entries, and premature treatment or study discontinuations, missing data is a common phenomenon in controlled clinical trials. With the widely accepted use of the intent-to-treat (ITT) analysis dataset as the primary analysis dataset for the analysis of controlled clinical trial data, the presence of missing data could lead to complicated data analysis strategies and subsequently to controversy in the interpretation of trial results. In this article, we review the mechanisms of missing data and some common approaches to analyzing missing data with an emphasis on study dropouts. We discuss the importance of understanding the reasons for study dropouts with ways to assess the mechanisms of missingness. Finally, we discuss the results of a comparative Monte Carlo investigation of the performance characteristics of commonly utilized statistical methods for the analysis of clinical trial data with dropouts. The methods investigated include the mixed effects model for repeated measurements (MMRM), weighted and unweighted generalized estimating equations (GEE) method for the available case data, multiple-imputation-based GEE (MI-GEE), complete case (CC) analysis of covariance (ANCOVA), and last observation carried forward (LOCF) ANCOVA. Simulation experiments for the repeated measures model with missing at random (MAR) dropout, under varying dropout rates and intrasubject correlation, show that the LOCF, ANCOVA, and weighted GEE methods perform poorly in terms of percent relative bias for estimating a difference in means effect, while the MI-GEE and weighted GEE methods both have less power for rejecting a zero difference in means hypothesis.

Estimating treatment effects from longitudinal clinical trial data with missing values: comparative analyses using different methods

Longitudinal pharmacometric models offer many advantages in the analysis of clinical trial data, but potentially inflated type I error and biased drug effect estimates, as a consequence of model misspecifications and multiple testing, are main drawbacks. In this work, we used real data to compare these aspects for a standard approach (STD) and a new one using mixture models, called individual model averaging (IMA). Placebo arm data sets were obtained from three clinical studies assessing ADAS-Cog scores, Likert pain scores, and seizure frequency. By randomly (1:1) assigning patients in the above data sets to “treatment” or “placebo,” we created data sets where any significant drug effect was known to be a false positive. Repeating the process of random assignment and analysis for significant drug effect many times (N = 1000) for each of the 40 to 66 placebo-drug model combinations, statistics of the type I error and drug effect bias were obtained. Across all models and the three data types, the type I error was (5th, 25th, 50th, 75th, 95th percentiles) 4.1, 11.4, 40.6, 100.0, 100.0 for STD, and 1.6, 3.5, 4.3, 5.0, 6.0 for IMA. IMA showed no bias in the drug effect estimates, whereas in STD bias was frequently present. In conclusion, STD is associated with inflated type I error and risk of biased drug effect estimates. IMA demonstrated controlled type I error and no bias.

The debate about what constitutes a valid analysis of clinical trial data is longstanding. While the intention-to-treat (ITT) principle seems to be widely accepted in the context of controlled clinical trials aiming to show superiority of an experimental treatment over a control, the best choice for a noninferiority trial is still under discussion. In this article, it is argued that the definition of analysis sets and the purpose of ITT and per-protocol analyses proposed in the International Conference on Harmonisation biostatistics guideline E9 should be revised to allow for more appropriate analyses, given that statistical methodology has been developed since the guideline was issued.

MSQ, the 16-item Migraine-Specific Quality-of-Life Questionnaire (Version 1.0), was developed by Glaxo Wellcome Inc. to assess the effect of migraine and its treatment on patients' health-related quality of life (HR-QOL). The MSQ was hypothesised to measure 3 meaningful dimensions: (i) Role Function-Restrictive; (ii) Role Function-Preventive; and (iii) Emotional Function. The objective of this research was to further investigate the number of dimensions as well as the items contained in each dimension through principal components factor analysis of clinical trial data. Secondary objectives were to determine whether the factor structure changed in post-treatment visits compared with screening visits, to make recommendations for coding the MSQ when the patient did not have a migraine in the previous 4 weeks, and to modify the MSQ if so indicated by this research. Results supported the existence of 3 distinct factors which agreed strongly with the hypothesised dimensions. The analysis of post-treatment data suggested that the underlying factor structure of the MSQ varies as a result of treatment. Based on evaluations of the 'did not have a migraine' response, it was concluded that it be dropped from the MSQ. All these changes have been incorporated into MSQ (Version 2.0) which is being evaluated in studies to determine if its psychometric properties are different than the properties of the previous version.

BackgroundTraditionally rheumatoid arthritis (RA) trials classify patients as responders and non-responders; they ignore the potential range of treatment responses. Group Based Trajectory Models (GBTMs) provide a more refined approach. They identify patient subgroups with similar outcome trajectories. We used GBTMs to classify patients into subgroups of varying responses and explore factors associated with different responses to intensive treatment in a secondary analysis of intensive treatment in the TITRATE clinical trial.MethodsThe TITRATE trial enrolled 335 patients with RA: 168 patients were randomised to receive intensive management, which comprised monthly assessments including measures of the disease activity score for 28 joints (DAS28), treatment escalation when patients were not responding sufficiently and psychosocial support; 163 of these patients completed the trial. We applied GBTMs to monthly DAS28 scores over one year to these patients who had received intensive management. The control group had standard care and were assessed every 6 months; they had too few DAS28 scores for applying GBTMs.ResultsGBTMs identified three distinct trajectories in the patients receiving intensive management: good (n = 40), moderate (n = 76) and poor (n = 47) responders. Baseline body mass index (BMI), disability, fatigue and depression levels were significantly different between trajectory groups. Few (10%) good responders were obese, compared to 38% of moderate, and 43% of poor responders (P = 0.002). Few (8%) good responders had depression, compared to 14% moderate responders, and 38% poor responders (P < 0.001). The key difference in treatments was using high-cost biologics, used in only 5% of good responders but 30% of moderate and 51% of poor responders (P < 0.001). Most good responders had endpoint remissions and low disability, pain, and fatigue scores; few poor responders achieved any favourable outcomes.ConclusionGBTMs identified three trajectories of disease activity progression in patients receiving intensive management for moderately active RA. Baseline variables like obesity and depression predicted different treatment responses. Few good responders needed biologic drugs; they responded to conventional DMARDs alone. GBTMs have the potential to facilitate precision medicine enabling patient-oriented treatment strategies based on key characteristics.RegistrationTITRATE Trial ISRCTN 70160382.

Achieving a target intraocular pressure (IOP) in patients with ocular hypertension and open-angle glaucoma has many barriers, some treatment specific and others patient specific. Nonadherence to IOP-lowering eyedrops is one of the major barriers to effective glaucoma treatment. There is a wide range of patients who may benefit from non-topical treatment modalities such as selective laser trabeculoplasty and sustained-release drug delivery, including those who face challenges with topical eyedrops, and those who find benefit in independence from daily medication administration. Sustained-release drug delivery options for lowering IOP include the bimatoprost intracameral implant 10 µg (Bim-I), a relatively new treatment option. This article provides expert opinion on the use of the implant based on available literature, supplementary analysis of clinical trial data, and the experience of the authors who have performed thousands of Bim-I administrations. A single administration of Bim-I has demonstrated the potential to reduce IOP and the need for topical IOP-lowering medication for up to 2 years. Evidence suggests that the safety profile of the implant is most favorable in patients with wide-open (Shaffer grade 4) angles. Recommendations regarding the administration procedure and patient follow-up are provided. The objective of this expert opinion is to help optimize patient selection for Bim-I administration, the implantation procedure, follow-up management, and clinical outcomes while minimizing and addressing potential complications.

Background Due to its potential nephrotoxicity, screening for pre-existing renal function disorders has become a routine clinical assessment for initiating Tenofovir diphosphate fumarate (TDF)-containing antiretroviral treatment (ART) or pre-exposure prophylaxis (PrEP) in pregnant and non-pregnant adults. We aimed to establish reference values for commonly used markers of renal function in healthy pregnant women of African origin. Methods Pregnant women ≥18 years, not living with HIV, and at 14–28 weeks gestation were enrolled in a PrEP clinical trial in Durban, South Africa between September 2017 and December 2019. Women were monitored 4-weekly during pregnancy until six months postpartum. We measured maternal weight and serum creatinine (sCr) at each visit and calculated creatinine clearance (CrCl) rates using the Cockcroft–Gault (CG) and Modification of Diet in Renal Disease (MDRD) formulae. Reference ranges for sCr and CrCl by CG and MDRD calculations were derived from the mean ± 2SD of values for pregnancy and postdelivery. Results Between 14-–and 40 weeks gestation, 249 African women not exposed to TDF-PrEP contributed a total of 1193 renal function values. Postdelivery, 207 of these women contributed to 800 renal function values. The normal reference range for sCr was 30–57 and 32–60 umol/l in the 2nd and 3rd trimesters of pregnancy. Normal reference ranges for CrCl using the MDRD calculation were 129–282 and 119–267 ml/min/1.73m2 for the 2nd and 3rd trimesters, respectively. Using the CG method of calculation, normal reference ranges for CrCl were 120—304 and 123–309 ml/min/1.73m2 for the 2nd and 3rd trimesters respectively. In comparison, the normal reference range for sCr, CrCl by MDRD and CG calculations postpartum was 40–77 umol/l, 92–201, and 90–238 ml/min/1.73m2, respectively. Conclusions In African women, the Upper Limit of Normal (ULN) for sCr in pregnancy is approximately 20% lower than 6 months postnatally. Inversely, the Lower Limit of Normal (LLN) for CrCl using either MDRD or CG equation is approximately 35% higher than 6 months postnatally. We provide normal reference ranges for sCr and CrCl for both methods of calculation and appropriate for the 2nd and 3rd trimesters of pregnancy in African women.