With encouragement from the U.S. Food and Drug Administration (FDA), using Patient Reported Outcomes (PRO) data to claim labeling became more and more popular. Well-defined and reliable PRO can be used to support a claim in medical product labeling. [1] It is found that there are an increasing number of regulatory submissions for new drugs to provide PRO data to support claims. DeMuro et al. (2013) [2] have reviewed drug approvals by both FDA and EMA for the years 2006–2010. They found that out of 75 drugs approved by the EMA, 35 (47%) had at least one PRO related claim approved by the EMA compared to 14 (19%) for the FDA.
What is a Patient Report Outcome and PRO instruments?
A patient reported outcome (PRO) is a health outcome directly reported by the patient who experienced it, without interpretation by physicians or others, about how they function or feel in relation to a health condition and its therapy. PRO could capture perceptive information of the patient, such as the symptom change of pain and itch. Instead of only focusing on physical measurements, PRO moves the focus towards how the patients feel, e.g. if they feel their life improved after treatment? PRO instruments (e.g., questionnaire items, instructions, and guidelines for scoring and interpretation) are used to measure these patient reports. [3] In today's environment we are seeing many electronic patient reported outcome (ePRO) solutions for clinical data capture available on the market and being integrated into study protocols.
What statisticians could contribute during the Development of a PRO Instrument?
The development of PRO instruments is a complex and iterative process. According to the guidance from FDA, there are 5 main aspects to develop a PRO instrument (as per the figure below). In this blog, we only discuss the area in which statisticians could contribute during the development process.
Reference: Guidance for Industry Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims, FDA, 2009.
1. Blinding and Randomization
According to FDA guidance “Open-label clinical trials, where patients and investigators are aware of assigned therapy, are rarely adequate to support labeling claims based on PRO instruments.” This is because the patient’s knowledge of which treatment they have received may influence their perception of their symptoms. For this reason, PRO endpoints should be avoided in studies where it is not possible for patients to remain blind to treatment allocation.
2. Checking the response options from the questionnaires
The options of the responses need to avoid potential ceiling or floor effects. During the process of PRO instrument development, if you found most of the patients response gathered with one option, you may want to check whether the question captures changes in patients. e.g. some questionnaires designed to measure patients may not able be sensitive enough to capture very mild symptoms. Due to this ceiling effect, the results can’t show any improvement, vice versa for floor effects. What’s more, you may also want to check if there are a sufficient number of responses. More responses may be required in order to capture worsening or improvement so that fewer patients respond at the top or bottom.
3. Reviewing the distribution of response options
In order to analyse responses, numerical scores will normally be assigned to each response option. Statisticians need to review if the scale of measurement is appropriate and also need to check the distribution of response options to ensure the numerical scores are appropriate. [1] If there is a wide enough range of responses and a large enough sample, then a normal distribution would be expected. If the patient responses are particularly skewed, it may indicate that the wording of the responses needs adjusting so that the responses are equally spaced in terms of increasing severity.
4. Handling Missing Data
As with any type of data, patients who withdraw from the trial with no PRO data could be problematic. Patients may withdraw due to dissatisfaction with the treatment, lack of efficacy, adverse event, over burden, unpleasant clinical trial experience, or the worsening symptoms of disease cause and be unable to visit the clinical centre. So, if the remaining data shows a positive treatment effect in terms of PRO, it may not be because the treatment provides an overall increase in patient satisfaction, it could be due to the patients who were dissatisfied having withdrawn from the trial.
The FDA suggests pre-specifying processes to get data on each patient at the time of early withdrawal from the clinical trial. If a measurement is taken at the time of withdrawal, this information can be handled according to rules established in the Statistical Analysis Plan (SAP). In clinical trials of terminal illnesses, it is critical to plan ahead for the event of missing data due to the potential death of subjects. Even with the best design, data still could be missing at the end of the trial. The SAP needs to explain plans for how the statistical analysis will handle missing data when estimating treatment advantage.
There are two types of missing data, the first type is missing a few items within domains, and second type is missing the entire domain. For the first type of missing data, the FDA suggests pre-defining the rule of how many items or the proportion of items allow to be missed, in order to consider the domain as sufficiently measured. This rule would need to be specified in the SAP.
There is no agreed best approach for handling the second type of missing, as all strategies contain strong or weak assumptions. The FDA recommends offering two or more sensitivity analyses with different methods for missing data imputation. Strategies of missing data imputation should consider the patient population, disease progression and burden. The strategies also need to be specified in the SAP.
5. Statistical analysis of PRO data
The cumulative distribution function (CDF) is an informative method to analyse the treatment benefit. We could examine the cumulative distribution function of responses between treatment groups to characterize the treatment effect and examine the possibility that the mean improvement reflects different responses in patient subsets. [1]
The above figure provides an example of a cumulative distribution curve. This technique has been used in product labeling to display treatment effect data. The figure shows that change from baseline ranges from a negative 6 to a positive 6 with positive changes indicating improvement in the PRO. Figure 2 shows the two drugs (A and B) are clearly distinguished from placebo beginning at the no change point to 6 points above. Different cumulative distribution curves can be anticipated depending on the distribution of the effect as measured by the end point and its variance. [4]
Conclusions
There are some advantages of using PRO to claim labeling, e.g. PRO could easily capture patients’ feelings. Since not all symptoms and impacts of medication can be detected from clinical tests. For some treatments for relief the symptom of pain and itch, PRO could track the changes of individual. Also PRO make patients feel more involved in clinical trials as their opinions are valued.
However, there are some disadvantages of PRO, e.g. the patient may forget to complete the PRO form, causing missing data. Additionally, the measurement in PRO is subjective, so the results could be influenced by other things going on for patients. What’s more, the PRO instrument takes a long time to develop, test and validate to make sure it is suitably sensitive and fit for purpose. It requires engaging FDA to discuss and review the PRO instrument before clinical trial protocols are finalised.
Thus, sponsors need balance the pros & cons of PRO before using PRO to support medical product labeling claim.
Learn more about our clinical biostatistics services, and how our statisticians could support you to design a defined and reliable Patient-Reported Outcomes (PRO) by scheduling a call with one of our sales representatives.
Reference
[1] Guidance for Industry Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims, FDA, 2009.
[2] DeMuro C, Clark M, Doward L, Evans E, Mordin M, Gnanasakthy A (2013). "Assessment of PRO label claims granted by the FDA as compared to the EMA (2006-2010)". Value in health. 16 (8): 1150–5. PMID 24326168. doi:10.1016/j.jval.2013.08.2293
[3] International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 1098 3015/07/S125 S125–S137
[4] Donald L. Patrick ,PhD, MSPH1 and etc. (2007), “Patient-Reported Outcomes to Support Medical Product Labeling Claims: FDA Perspective”. Value in health. Volume 10, Supplement2.
