Synthesis of quality of life data in systematic reviews: problems and proposals

Tags: Oral
Grosselfinger R, Scheibler F, Lange S

Background: The Institute for Quality and Efficiency in Health Care (IQWiG) prepared a systematic review on interstitial brachytherapy in localised prostate cancer. Within the framework of this review, the comparison of quality of life data obtained from different studies was problematical for the following reasons: Different studies applied different instruments to measure the same outcome. If the same instruments were used, results from different subscales were reported. In addition, the interpretation of results was further hampered by different effect measures, ranges of scales or directions of score changes.

Objectives: We suggest a methodology to transform results from different quality of life instruments in order to compare outcomes that would otherwise not have been comparable.

Methods: The following methodological approach in particular applies if a meta-analysis is not feasible or not planned. Quality of life data were processed in a three-step approach: (1) summarisation of scales of different instruments into newly-defined categories that reflect similar content; (2) summarisation of the different effect measures into five categories (which also described the direction of the effect), using reported or self-calculated p-values; (3) summarisation of the transformed results in a multidimensional table spanning all instruments.

Results: The IQWiG review included 5 studies referring to results from 14 different quality of life instruments with 54 scales reported. The application of strict methodological criteria would only have allowed the comparison of results from a single scale, which was used consistently in 3 of the 5 studies. After the described transformation of data, results from 22 scales (summarised into 9 categories) were comparable.

Conclusions: Even though a considerable amount of data was available, if we had adhered to the requirement of comparing only data from identical instruments, we would not have been able to summarise results. In order to identify possible trends, a loss of detail was accepted by comparing data from similar instruments in a semiquantitative manner. The loss of information associated with the type of data transformation applied led to an increase in comparability and clarity, which can be more useful than the purist description of unmanipulated original data.