“Why would pharma need epidemiology skills?”
I’ve been getting lots of questions from current epidemiology students on how to use their skills in industry, and today I’d like to discuss how we can use epidemiological skills in drug development.
During my graduate studies, I was introduced to clinical trials as a distinct field, separate from epidemiology and observational studies. As a student, I didn’t realize that these methods could be integrated and used together. The introduction of the 21st Century Cures Act of 2016 solidified not only the utility but the necessity of this integration. This has opened up new avenues for epidemiologists, allowing them to move beyond traditional roles in government, academia, and post-approval surveillance to contribute to pharmaceutical drug development and approval processes.
Understanding the Terminology “Real World Data”
It wasn’t mentioned during my time in school, but epidemiologists have long been involved in the pharmaceutical industry. However, what we referred to as observational or community data in epidemiology is known as “real world data” (RWD) in the pharmaceutical research world. RWD encompasses all data not derived from clinical trials, including administrative datasets, registries, community surveys, and surveillance databases.
Traditionally, RWD has been used in Phase IV trials, or post-marketing trials, which are crucial for assessing patient safety and accessibility following drug regulatory approvals. However, with the introduction of the 21st Century Cures Act and considering the high costs of clinical trials—where moving from Phase 1 to 3 can take 10-15 years and a Phase 3 trial costs, on average, $225 million—RWD’s role has become significant even before a drug is approved.
This integration of epidemiological skills and real-world data is reshaping drug development. It allows epidemiologists to extend their impact beyond traditional roles in government, academia, and post-approval surveillance, contributing significantly to pharmaceutical drug development and approval processes.
What Epidemiologists Bring to the Table
RWD is inherently complex and messy, which traditionally limited its use in randomized clinical trials (RCTs). Unlike data collected specifically for research purposes, RWD is often not standardized and is fraught with numerous known and unknown confounders.
Epidemiologists are adept at dealing with the intricacies of RWD. Being trained to work with messy data means that with every new dataset, they are anticipating variability in data quality, looking for gaps in data collection, critically appraising the validity of data sources, and assessing the presence of both measurable and hidden confounders.
Supporting Phase I to III Trials:
Patient Recruitment and Trial Feasibility
(Brøgger-Mikkelsen M, Ali Z, Zibert JR, Andersen AD, Thomsen SF. Online Patient Recruitment in Clinical Trials: Systematic Review and Meta-Analysis. J Med Internet Res. 2020 Nov 4;22(11):e22179. doi: 10.2196/22179. PMID: 33146627; PMCID: PMC7673977.)
Patient recruitment is costly, and as research move towards rare disease, it can be a challenge to not only find patients, but representative cohorts. RWD can significantly enhance this process thereby increasing the feasibility of clinical trials.
By analyzing extensive healthcare databases, researchers can quickly identify and target potential participants who precisely fit the trial’s inclusion criteria. This approach not only accelerates the recruitment process but also improves the likelihood of enrolling participants who are truly representative of the condition under study.
RWD can also be used to provide a realistic estimate of the number of eligible participants within a given geographical area or health system, helping to assess the feasibility of the trial. This includes determining if there are sufficient participants to support robust study conclusions and whether recruitment goals are achievable within the desired timelines.
Protocol Optimization
RWD also plays a crucial role in optimizing clinical trial protocols. For example, by leveraging EHR data, epidemiologists can gain detailed insights into medication use in actual practice, and understand the interplay of comorbid conditions (i.e. confounding variables) within a target population. This includes understanding which medications are most commonly prescribed and their outcomes, which can guide the inclusion of the most relevant and widely used treatments in the trial.
This knowledge can then be used to help inform the selection of medications for different trial arms. By understanding the nuances of medication performance and patient responses in diverse subgroups, researchers can design trial arms that are both effective and safe. This process includes predicting potential adverse events that might not be apparent in less heterogeneous populations, allowing for more targeted and precautionary measures to be integrated into the trial design.
This information can also help in tailoring dosing regimens and treatment durations to better match the conditions and practices observed in actual clinical settings, thereby increasing the generalizability and applicability of the trial outcomes to broader patient populations.
Enhancing Statistical power and Design Adjustments
RWD is invaluable in enhancing the statistical power of clinical trials and facilitating necessary design adjustments. By providing access to extensive pre-existing data, epidemiologists can conduct preliminary analyses to better estimate effect sizes, identify potential confounders, and anticipate variability in treatment responses. This preliminary insight allows for the design of trials that are statistically robust and adequately powered to detect meaningful effects, minimizing the risk of Type II errors. In addition, RWD enables adaptive trial designs where modifications to the trial protocol, such as sample size adjustments or changes in the primary endpoints, can be made in response to interim data.
External Control Arms
Using RWD to create external control arms is very useful particularly when traditional randomized controls are not feasible due to ethical concerns, rarity of the condition, or logistical constraints. An external control arm is constructed using data from patients who have previously undergone either standard treatment or no treatment, as recorded in RWD or previous study databases. This approach allows epidemiologists to compare the outcomes of a new intervention against a well-defined external comparator group without the need to recruit, consent, and follow a separate control cohort concurrently. This not only speeds up the trial process but also reduces costs significantly. Furthermore, external control arms can enhance the ethical dimensions of clinical research by reducing the number of patients exposed to potentially less effective treatments.
Supporting Phase II to IV Trials:
Adjusting for Confounders
Even though, in theory, the controlled environment of RCTs is designed to adjust for confounders via randomization, there may be scenarios where more advanced methods like stratification combined with propensity score matching (PSM) and multivariate regression modelling are necessary. This need typically arises in instances of imperfect randomization, where unforeseen imbalances in baseline characteristics across treatment groups occur. Let’s talk a bit about these methods in the context of clinical trials.
Stratification
Stratification is used in both the design and analysis phases of clinical trials to ensure that subgroup variables, which could influence outcomes, are evenly distributed across treatment groups. This involves categorizing participants based on key characteristics like age, gender, or disease severity before randomization, allowing for control over these variables from the outset. During the analysis phase, stratification offers more precise estimates of treatment effects within each subgroup, improving the reliability and applicability of the results across different population segments.
Propensity Score Matching
After stratification Propensity Score Matching (PSM) can be used to further refine the equivalence between treatment and control groups. PSM is a statistical technique designed to estimate the effect of an intervention while accounting for the covariates that influence the likelihood of receiving the treatment.
In practice, PSM begins by calculating a propensity score for each participant, which represents the probability of receiving the treatment based on observed covariates. These covariates might include variables such as age that could affect both the likelihood of receiving the treatment and the outcomes of interest.
Once these scores are calculated, participants in the treatment group are paired with participants in the control group who have similar propensity scores. Having similar propensity scores implies that they have comparable backgrounds in terms of the observed covariates. Therefore we have a treatment and control group that are statistically similar on all observed covariates, essentially creating a “pseudo-randomized” dataset.
This matching process helps to simulate the conditions of a randomized controlled trial (RCT). In an RCT, participants are randomly assigned to either the treatment or control group, which theoretically balances all known and unknown covariates across the groups. While PSM does not account for unobserved variables, it significantly reduces the impact of measurable confounding variables to enhance the ability for causal inference.
Integration of PSM with Multivariate Regression Modelling
Following the matching process, multivariate regression modelling is employed to analyze the outcome variable of interest. This step is crucial as it allows for the adjustment of potential residual confounding that might still exist after matching. In multivariate regression, multiple covariates are included simultaneously in the model to estimate the treatment effect while controlling for other variables that could influence the outcome. This helps in understanding the relationships between variables and in distinguishing the specific effect of the treatment from other influencing factors.
The combination of these two methods leverages the strengths of both. PSM ensures that the groups being compared are balanced on many covariates, reducing the bias due to non-random treatment assignment. Then, multivariate regression deals with the complexity of multiple interrelated variables, providing a nuanced analysis of the data. This dual approach not only increases the credibility of the findings by minimizing bias but also enhances the generalizability of the results to a broader population.
Clinical Trial Emulation Using ITT Approaches in Different Phases
Epidemiologists employ techniques to emulate intention-to-treat (ITT) analyses, a cornerstone of RCTs, across various phases of clinical trials using observational studies. ITT analysis includes all participants as originally assigned to treatment groups, regardless of adherence to the intervention protocol. This method maintains the benefits of randomization by providing an unbiased estimate of the treatment effect, crucial for the validity of trial outcomes.
Application in Phase II and III Trials
In Phase II and III trials, ITT emulation is particularly beneficial for assessing the efficacy and safety of a treatment under conditions that closely resemble real-world usage. By applying ITT principles using RWD, researchers can ensure that the study reflects a comprehensive view of the treatment’s effects across a diverse population, including those who may not fully adhere to the protocol. This approach can leverage advanced techniques such as propensity score matching to ensure that the comparison groups are well-balanced in terms of baseline characteristics, enhancing the reliability of the efficacy and safety assessments.
Extension to Phase IV Trials
In Phase IV, ITT emulation continues to play a critical role by examining long-term outcomes and potential adverse effects after the drug has been marketed. It allows for the inclusion of a broader patient population that reflects varied adherence patterns seen in routine clinical practice. Here, the use of multivariate regression modelling alongside ITT principles helps in adjusting for confounders that emerge due to differences in treatment adherence and patient characteristics, providing a detailed understanding of the drug’s performance in a real-world setting.
Looking Forward
As we progress, epidemiologists are at the forefront of innovating and refining methodologies to harness the potential of Real World Data (RWD). By developing and applying advanced statistical models, such as mixed-effects models and machine learning algorithms, we are better equipped to manage the complexities and irregularities of large datasets. These sophisticated epidemiological skills are instrumental in transforming unstructured RWD into meaningful, actionable insights that significantly enhance decision-making in drug development and beyond.
The integration of RWD in clinical trial teams is not without challenges, yet the opportunities it presents are immense. While there remains skepticism among some traditional clinical scientists about the utility of RWD due to its inherent “messiness”, it’s important to recognize that RWD complements randomized controlled trials (RCTs), which are the gold standard in drug development. RWD is not about replacing RCTs but enhancing them, helping to overcome obstacles and accelerating the development of drugs.
With the vast data sources available today and the evolving technology, we can bridge the gap between RCT and RWD to expedite drug development and deliver treatments faster to those who need them most.
To My Fellow Epidemiology Students/Colleagues
Your passion for epidemiology and your skills are crucial for advocating the integration of RWD. The value of RWD will only continue to grow, and it’s essential not to let skepticism deter you from pushing the boundaries of what’s possible in drug development research. As you think about your next moves, know that your skills are applicable and in demand in both the public and private sectors.
Leave a Reply