Just a moment, the page is loading...

Selection of comparable subjects from different treatment groups when randomization is not feasible

Selection of comparable subjects from different treatment groups when randomization is not feasible


Business Transformation Solutions Global Business Solutions Novartis Healthcare Pvt. Ltd. Salarpuria-Sattva Knowledge City Raidurg, Rangareddy District Madhapur / Hyderabad, Rangareddy 500008

13 November 2023

Whenever a researcher needs to assess treatment effect (mathematical comparison) for 2 or more treatments to get unbiased results, the randomization technique (suggested by RA Fisher) has been adopted as the Gold Standard. Randomization is a process of assigning the treatment to subjects, assuming that each subject has equal chance to participate in any treatment group. It is assumed that random allocation of subjects to treated and comparator groups approximately balances on subject's background, but in the following situation researchers need to look for alternative methods for randomization.1. Randomization (true experimental designs are not always possible, practical, logistical, even desirable and other reasons) possible every time or 2. Even if randomized study, a) Randomization broken orb) Randomization failed to recruit similar profile of subjects (it can be said as unethical randomization) orc) Another treatment (new) group difference estimate expected to find or3. Treatment effect in real-word evidence (observational) study. Studies where the objective can be answered by data collected thorough observational systems, as they operate in normal (routine) practice without any interventions, implementation of randomization assignment rule, can be considered as real word evidence studies. There is no requirement to conduct separate studies to answer specified objectives.Drawing conclusions in such non-randomized studies is challenging, and it's encouraging for the statistician to look for another method, which can give identical results as that from a randomized trial.The Propensity Score Matching (PSM) method can be used for this purpose. PSM is a statistical technique, which can be used to select comparable subjects from treatment groups based on the subject's background data. There is lot of discussion on this topic and statisticians have not reached any consensus. Statistician (Rosenbaum & Rubin, 1883) and econometrician (Heckman, 1978, 1979) made a substantial contribution by developing and refining new approaches for estimation of treatment effect from observational data. Collectively these approaches are known as propensity score analysis. PSM may be helpful when randomization fails or is impossible (Barth et al, 2007).If properties (individual background) of subjects at baseline are as similar as possible in comparable treatment groups then it can be said that data is comparable and whatever differences between treatments are observed, it's due to treatment group, not by other known/unknown factor. An example of an observational study. Suppose there are two treatment groups A and B with 23 and 186 subjects, respectively. From this information, it cannot be sure that. 1) Study design is balanced and have comparable data. 2) Treatment groups are adjusted with covariates (other factors that can affect the outcome).3) Assurance that result is unbiased.4) Presence of estimated difference is because of treatment groups only, and not other known/unknown factor (demographic and baseline characteristic).Paul R ROSENBAUM and Donald B. Rubin suggested central role of propensity score in observational studies for causal effects. The propensity score matching technique helps to find pair of subjects from two treatment groups which are not identical but are as similar as possible. Pair means, if one subject chosen from low treatment group (in which subjects are fewer) then a particular subject having similar properties from high treatment group (in which subjects are higher), if a pair from treated and comparator subjects have similar propensity score are considered comparable, even though they may differ on values of specific covariate. The key question of interest is how to choose such pairs? How to define pairs are “similar”? If there are many such pairs then the comparison between two treatment groups may be fair. It will be close to a randomized trial. These matched groups of subjects can be used to compare the two treatments. This work aims to solve the problem of optimal comparison of treatment groups using profile matching methods when randomization is not possible/unethical.

[{ "PostingID": 21103, "Title": "NOVARTIS-CENA713D2344", "Description": "A 24-Week Efficacy, Safety and Tolerability of Rivastigmine Patch Study in Patients With Probable Alzheimer's Disease" }]

Statistical Analysis Plan