Can Patient Chart Pulls Help Predict Physician Behavior?

Welcome back to my three-part series on physician treatment behavior. If you missed the first installment, you can find it here. In this second installment, I will be discussing the use of patient chart pulls as a way to understand both current and future physician prescribing.

Patient chart pulls offer significant advantages to allocation studies, but come with their own set of weaknesses. In a patient chart pull, physicians bring actual patient charts into the study and relay information about their patients and their treatments to researchers. One benefit of a patient chart pull is that they can offer a wealth of information about real-life patients. Researchers can look at physician treatment decisions across any of the patient characteristics collected.

Cumbersome to Collect

The downside to getting such detailed information about patients is that it can be extremely cumbersome data to collect, requiring extensive time spent on patient characteristics, current treatments, and possibly future treatments. Because it takes so long to collect this data, fewer patients can be analyzed compared to other methodologies (assuming the same length of interview). Having fewer patients means wider confidence intervals around estimates, less power to detect differences between shares, and less power to detect if patient characteristics drive treatment decisions. There is also the possibility of data entry mistakes during interviews, as this is an exceptionally tedious task. All of this is on top of the issues associated with handling this type of data such as HIPPA compliance, Adverse Event compliance, etc. Collecting so much information is also boring for the respondent and can quickly lead to fatigue, which can also contribute to less accurate results. All of this leads to lower levels of confidence in results and fewer insights generated from the data collected (read: less return on investment into market research).

Uncertainty of Sample Patients

Another potential down side to chart pulls is that the researchers are subject to what I’ll call the “gods of randomization”. The assumption is that physicians are bringing a perfectly random sample of their breast cancer patients (sticking to my original example) into the study. However, this is a large assumption. Some studies may get the target sample of physicians or even patients into the study, but this does not solve the problem. The problem is that patient characteristics may not be fully represented or uncorrelated. Let me give you an example of how this could become a problem. Say for example, that weight and stage of disease are correlated with one another and drive physician decision to treat with a new drug. These two variables increase and decrease together – they’re correlated. This is a problem if your business objective is to understand which patients that physicians think are the best fit for your new medication because you can’t pull apart if that choice is based on weight or stage of disease.

Patient chart pulls are most useful for understanding current physician treatment decisions. These studies can answer questions about which patients are treated with which medications and when. However, getting this information can be costly both in terms of time, money, and statistical power. Also, there is no guarantee that the distribution of patients included in these studies perfectly matches real-world patients (even if steps have been taken to get a representative sample of physicians).

Please stay tuned for my next and final blog post in this series where I will examine patient simulation studies…