Meta-analysis is not always the best way to round out a systematic review: a few thoughts prompted by the COVID-19 pandemic and spiced-up with an earthquake

摘要

The last few days of March 2020 in Zagreb, Croatia, were anything but usual early spring days. By the end of February, first patients with COVID-19 infection were identified, and hospitals were preparing for the (expected) increased number of patients, while most of them were considerably damaged by two strong earthquakes that hit Zagreb on March 22. Just about the end of March, a paper was published ( ) that drew our attention – we considered that it might be useful to forward it to our hospital colleagues who did not have time to search for the literature that might guide their practice. A collaboration of several research groups resulted in a prompt, thorough, and up-to-date (at the time) systematic review focused on observational studies reporting on clinical, epidemiological, laboratory, and radiological characteristics and disease severity and course in COVID-19 patients ( ). A thorough risk of bias assessment was performed using a tool adapted for this kind of studies. A total of 60 studies were finally included – 20 case reports, 37 case series, and 3 epidemiological reports involving between 1 and close to 50 000 patients per study, mostly from China, but also from 10 other countries ( ). A number of meta-analytical estimates were generated in order to assess the prevalence of individual symptoms/signs, laboratory test values, and mortality – however, all were so severely heterogeneous that were completely non-informative. On the other hand, simply summarized data (as simple raw proportions), such as the percentage of patients with a certain laboratory value within or outside the physiological range, and narrative parts on certain findings were more informative ( ).