15283-B-Blokker

62 Chapter 4 Due to insufficient data, we were also unable to test whether the agreement percentages on cause of death were any better than chance, since, with incomplete 2x2 tables, the chance-corrected proportional agreement ( g -statistic) could not be correctly calculated. Also, variability of the investigated study groups and study methods, and the information that was reported in the articles was too large to combine study outcomes in a meta-analysis. For example, in studies using radiological imaging, one, two, four or six (specialized) radiologists reviewed the images. Previous experience in post-mortem imaging was mentioned in 6 studies: it varied from no experience to 5 years of experience, and was not comparable between studies. Though Roberts et al. 110 found that previous experience did not result in more correctly diagnosed causes of death. Moreover, only two studies calculated an inter-observer agreement (kappa): Weustink et al. 57 reported kappas of 0.85 for CT and 0.84 for MRI, and Ross et al. 108 reported a kappa of 0.94. In addition, when comparing a new method to the reference standard in a validation study, the investigators performing one method should ideally be blind to outcomes of the other. This might not have been the case in eight of the studies reviewed, in which the agreement or sensitivity percentages may have been influenced, possibly biasing their value. Controversially, blinding induces failure to detect false positive and false negative results. Christe et al. 124 reported that both gas and fractures were better detected at imaging than autopsy. To prevent these imaging findings from being registered as false positives, they had the findings confirmed after a second look at the autopsy. The same way, taking a second look at the radiologic images after autopsy could rectify false negative results. However, in both situations the findings were not originally reported, so they may be missed again in the future. Another limitation, which is almost inevitable due to the kind of studies investigated, is knowledge of the medical histories prior to performing autopsy. None of the reviewed studies reported, that the investigators of conventional autopsy and its potential alternative were uninformed about the case circumstances. Therefore, this prior knowledge may have influenced the outcomes of agreement between the two methods, for known pathologies are more likely to be identified than unknown ones.