By Brian Hull, M.B.A., President, GfK Strategic Marketing
Product teams understand how specific patient populations flow through various points in the medical system. Auditing patient charts allows us to measure referral, diagnostic, treatment, fulfillment and follow-up patterns. Marketing teams can then identify leverage points in each phase to exploit in future launch strategies. Postlaunch, patient chart audits are extremely useful in helping product teams understand how their drug and competitive agents are truly being used. This acts as a catalyst for investigating the accomplishments of current strategies and adjusting programs to maximize success.
We are constantly reminded in today’s economic environment that pharmaceutical product teams struggle with maximizing their portfolio’s success while minimizing the expensive risks associated with an infinite number of business decisions. Consequently, product teams should look to primary market research that offers the versatility to answer a large number of tactical and strategic questions. The creative use of patient chart audit research fits the bill.
Patient chart audits as a primary research methodology can enhance a traditional market research analysis by yielding more than just a basic understanding of who prescribes what drug for whom and when. The ultimate is to properly link specific patient data with physician attitudinal and behavioral information because that is what is really required for effectual marketing. Patient chart audit studies can drive positioning strategies, assess why a drug is being treated as a “switch to” or a “switch from” agent, determine what triggers decisions throughout treatment plans for various patient populations, and even predict behavior through the creation of a patient-records choice model. However, patient chart audit studies can only reach their true potential through creative study design, excellent execution, innovative analysis, and imaginative presentation of detail-rich data.
Prior to engaging in patient chart audit research, it is important to recognize the many hurdles that must be overcome to ensure the final dataset is rich, accurate, complete, and prepared for both simple and complex analyses. One must never deviate from the fundamentals when conducting patient chart audit research. Key success factors include sampling sophistication, strong field management, and extensive experience designing and executing patient chart audit studies.
For traditional market research studies, devising sample designs tends to be quick and easy. A brief conversation with an experienced researcher or a methodologist is enough to confirm a predetermined sample design. This is not the case for patient chart audit research, where generating a proper and sensible sampling plan takes time and effort. Clever and purposeful sample plans are needed to ensure the results can be projected to the larger population of patients. Ample consideration must be paid to sampling error management to achieve high confidence levels for the results.
A solid sampling plan ensures the patient chart sample provides adequate representation and precision. By sampling plan, we mean not only figuring out the sample design for the physicians, but also determining how many and what type of patient charts to ask physicians to produce. Patient chart samples have multiple stages, with each stage having its own design. Physicians are sampled in the first stage. The physician sample design details how many physicians to include in the study, how to stratify those physicians, and how many physicians to recruit per stratum. Patient charts are sampled in the second stage. The patient-level design identifies how many patient records to collect per physician, how to stratify those records by patient type, and how many patient records to collect per patient type.
Many suggest trying to save money by decreasing the number of physicians who participate and compensate by asking each one to produce a higher number of patient records. This may seem like a reasonable compromise, but there are limits. The problem of most concern is that patient charts produced by the same physicians are more similar than patient charts produced by different physicians. In effect, asking each physician to produce a greater number of patient records causes intracluster correlation, which decreases the sampling efficiency. When asking individual physicians to produce their patient charts, it is often better to have them do so across a variety of different patient-type segments than to produce charts of patients who are too similar to each other.
For example, if we are interested in collecting patient chart data for mild and moderate-to-severe patients, it is better to have each physician provide data for both severity segments as opposed to having one set of physicians provide data for mild patients and one set to provide data for moderate-to-severe patients. Generally, samples with smaller numbers of physicians and more patient records per physician will cost less, but the sample with more physicians (and the same overall number of records) will produce more precise estimates in most cases. As a result, designing the sample requires the need to balance the cost of collecting the data with the level of precision desired in the final analysis.
Patient chart audit research is often used to provide estimates of difficult-to-size patient segments, either because accurate secondary data does not exist along patient dimensions of interest or because physicians cannot reliably tell us how many patients they have in specific patient subgroups. Relatively simple sampling approaches like asking physicians to provide data on their “last or next five” patients of interest might seem attractive but are prone to oversampling patients who frequently visit their doctor. We could ask physicians to provide data from charts that fit detailed patient characteristics, but this may complicate the task too much and physicians may ultimately shy away from participating. In such a scenario, we should get creative and consider doing something unconventional to accurately size the market. An inventive and useful way to overcome the hurdle of projectability is to implement the following two-step data collection process.
Step 1: Have physicians keep a diary of patient visits for a couple of weeks and fill out a survey to capture practice characteristics and attitudinal issues.
- An experienced methodologist can take the patient visit diary data and estimate patient-type frequencies overall and within practices.
- The risk of oversampling patients who frequently visit their doctor is managed by collecting the frequency of visit information on each record, and using it when weighting the patient record data.
- This information is used to develop a detailed sampling plan of physicians and patient charts. An important result is that in the next step, each physician can be asked to provide information on patient charts tailored to his/her individual practice composition.
Step 2: Physicians are asked to provide patient charts for a subset of patient identified in the diary.
- Linking physician ID numbers to patient charts allows for an appropriate weighting scheme and the ability to tie results to practice characteristics.
If the patient sampling dimensions are simple, accurate secondary data on the size of the patient segments may be available or physicians may be able to provide accurate estimates. If that is not the case, then creative use of the aforementioned two-step process will allow the study results to be reliably projected to the overall patient population at large.
Strong field management is necessary for the successful completion of patient chart audit research because the logistical difficulties can quickly get out of control. The crux of the data collection process consists of physicians, likely with the assistance of their nurses, pulling patient charts and transferring information from each one to a questionnaire. This requires a level of trust and commitment above and beyond the participation in a typical research study. It is very helpful to have seasoned market research professionals continuously address issues along the way and shepherd data collection to the end.
Extensive experience conducting patient chart audit research is the most important success factor. Practice and familiarity are the only characteristics that can guarantee adequate coverage of all the nitty-gritty details necessary to effectively design and execute patient chart audit studies.
Creative design of the patient record form to ensure it collects the “whys” and attitudes that explain the behavior detailed in each patient chart can yield data to drive positioning strategies. A sample outline of an actual patient record form, which demonstrates the type of valuable information that can be retrieved to offer rewarding marketing recommendations, is shown below.
Outline of Patient Record Form
- Patient demographics and characteristics (e.g., age, gender, employment status, ethnicity, educational status, living circumstances, prescription drug coverage status, comorbidities, level of involvement in treatment decisions)
- Evaluation and diagnosis (referral pattern, who made initial diagnosis, timing of diagnosis, symptoms, triggers, tests, factors preventing an earlier diagnosis if appropriate, severity and functional status at time of diagnosis)
- Treatment history
- First treatment – prescribing physician, rationale for choice, goals of treatment, initial dosing, dose adjustments, attribute perceptions, satisfaction with treatment, change in severity and functional status
- Repeat for all subsequent treatments (all lines of therapy) and obtain rationale for all changes and switches
- Current severity and functional status
- Fulfillment (how patient acquires treatment, perceived level of compliance with each treatment and rationale)
- Follow-up (original visitation frequency and rationale, change in visitation frequency over time and rationale)
- Assessment of prognosis (chance for improvement)
Exhaustive analysis of the data above from actual patient records will help drive positioning strategies by answering such questions as:
- What is the derived importance of product attributes that truly drive therapeutic decisions in the real world?
- Are patients involved in the treatment decision and do their attitudes matter?
- What physician specialties are prescribing which drugs and why?
- Which drugs are “switch from” agents and why?
- Which drugs are “switch to” agents and why?
- Who is receptive to newer agents?
- What are the predictors for using certain classes of agents?
- What is the patient-type positioning for current treatments?
- What is the most likely patient-type positioning for a new agent based on how its profile compares to the competition?
- Where are the gaps and opportunities in the market in terms of which patient types are being underserved?
- How are treatments positioned along clinical and nonclinical attributes?
- What are the unmet needs in the market?
- What types of associations does a new product need to be considered first line, second line or later?
In addition to univariate and bivariate reporting of the descriptive detail-rich patient chart data, multivariate reporting can be very useful. For example, a CART (classification and regression trees) analysis can be conducted to identify the key drivers for why physicians choose a specific agent over other agents in the same category. The following graphic shows an example of the output of a CART analysis in the migraine treatment area.
An inventive way to maximize the information potential of patient chart audit research is to create a predictive choice model from the actual patient record data. Building a choice model from actual patient record data can avoid the shortcomings of typical patient conjoint models because: 1) data from real world patient records are more valid and rich than data assigned to hypothetical patient profiles; 2) respondents who participate in typical patient conjoint model studies can experience fatigue when completing the survey which results in lower-quality data; and 3) patient conjoint models are potentially more likely to leave out an important patient characteristic while the patient record form in chart audit research can be much more exhaustive.
Creating a predictive choice model from actual patient record data would be of great value to all marketing teams that want answers to a wide variety of what-if scenarios. The choice model estimates the relative influence of different characteristics on treatment choice. These importance measures are stable enough to provide reliable estimates of how the influence of different patient characteristics may change over time. The model would also predict the influence of patient characteristics on treatment choice. A key deliverable is a patient simulator. The simulator allows the user to conduct a preference share analysis for endless patient-type scenarios. The simulator is in an easy-to-use Excel format where different scenarios are run by changing various patient characteristics. The following graphic is an example of a patient choice model simulator in the chronic hepatitis B treatment area.
Combining creative determination with a commitment to fundamentals significantly increases the impact of patient chart audit research on product teams’ marketing strategies. The way to achieve this is to look beyond the inherent value of patient record data and think of innovative ways to apply sophisticated research techniques for the analysis. A patient-chart database of statistics based on thousands of patient charts clearly answers the questions about who, what, when, and how, but we can only be successful at answering questions about why and what to expect in the future by creatively linking those statistics to other behavioral and attitudinal information.
Beyond Traditional Market Analysis: Creatively Using Patient Chart Audit Data
