Estimating the number of patients exposed to a particular drug is an important part of safety signal detection, but Vimal Pannala and Partha Chakraborty, from IT firm Cognizant, argue that
industry needs to adopt better standard definitions and automated processes to make full use of such data
The importance of safety signal detection - the rapid identification of potential unwanted side-effects has grown in recent years due to the increasing number of high profile withdrawals and the late-stage failure of drug candidates, as part of the ongoing scrutiny of drug safety.
In safety detection it is important to consider not just the adverse event (AE) reporting rates, but also the number of patients who might have been exposed to the drug (patient exposure). While AE reporting rates may keep rising month-on-month, it becomes a potential safety signal only when this increase is not explained by a proportional increase in patient exposure.
Patient exposure data is also required for reporting in Periodic Safety Update Reports (PSURs) (Section 2.5 of the PSUR) and other aggregate, cumulative reports capturing the safety profile of a drug, submitted in International Council of Harmonisation (ICH) regions.
Despite these requirements, there is a lack of industry standard definitions, which can have implications for the pharmaceutical industry.
While patient exposure estimation has been a regulatory requirement for some time, it is not regarded with the same level of seriousness as AE reporting rates. The following pointssummarise the current state of patient exposure estimation within most pharmaceutical companies:
- There is a lack of industry standard definitions and terminologies that would allow a more structured approach for patient exposure estimation;
- Estimation of patient exposure is a completely manual process in most pharmaceutical companies; and
- It is estimated only for the purpose of its inclusion in PSURs and its use in signal detection is minimal. This may lead to false positives that could be avoided if patient exposure is accurately estimated and used in signal management.
Data for patient exposure estimation for post-marketing comes from three sources:
1. Third-party prescription databases (such as IMS);
2. Primary sales data (from the company's financial database that would record £ value of sales); and
3. Bulk drug sales to a contract manufacturer.
Capturing data from these sources poses its own challenges. While prescription data is often the most accurate and has the smallest time lag between the time of data collection (at prescription) and end user consumption, such data may not always be available or is often incomplete (e.g. not all major retailers are covered). Furthermore, it can be very expensive.
When primary sales data is used, the available information may not account for free samples, product returns or unsold inventories lying within the retailer. There is also a significant time lag between the time of data collection (at primary sales level) to the final sale to the end consumer, and end user consumption.
Bulk drug sales data is often the most inaccurate measure of patient exposure, as there is wastage of active substance during drug manufacturing that is not accounted for when calculating the number of drug units manufactured. This data has the greatest time lag between the time of data collection (during bulk drug sales) and end user consumption.
There are ways of dealing with the challenges in patient exposure estimation. For example, the time lag between the time of data collection (during prescription, primary sales or bulk drug sales) and final user consumption introduces an error in patient exposure estimation as data from these sources could be several months old. To overcome this, most companies extrapolate available information based on known sales/prescription trends. However, this extrapolation may not be totally accurate and some statistical error does creep into patient exposure estimation.
Patient exposure is often very difficult to estimate accurately in terms of number of patients exposed and hence is normally estimated in person-time. This is because patient exposure in person-months or person-years is easier to identify from available, industry standard data sources if annual or monthly per capita intake of a drug is known.
However, in some cases (especially for single-dose drugs), patient exposure may be expressed in terms of persons exposed, whereas in the case of non-chronic drug treatments (e.g. anti-biotics) it is expressed as treatment courses. But in the case of over-the-counter drugs, where it is difficult to predict the usage pattern, it is sometimes expressed in net sales.
Pharmaceutical companies sell their drugs around the world in different ways and through a variety of channels. For example:
- under multiple trade names
- through multiple partners
- in multiple formulations
- in multiple strengths
- in multiple stock keeping units
All of these add to the complexity in accurately estimating patient exposure.
There is significant scope to automate this process by integrating required data from multiple sources such as sales, financial, manufacturing and prescription databases, into a patient exposure data warehouse.
This would allow for standardisation of patient exposure estimation for each drug as a pre-built report within the data warehouse.
Another option would be to integrate patient exposure data into a global safety data warehouse, and automate signal detection further by using patient exposure estimates generated from pre-built reports (as outlined above) with AE reporting rate trends to determine if a signal has been generated.