Reply to the letter regarding 'A systematic review of hospitalization resulting from medicine related problems in adult patients'.

Abstract

We are grateful to Mrs Souza, Dr Saez-Benito, Dr Correr and Dr Fernandez-Llimos for their interest and thoughtful comments on our review 1. Regarding the exclusion criteria, we decided to exclude non-English studies in order to avoid bias from translation. Thus, bias could be introduced due to the existing chaos surrounding the definition of MRPs (189 definitions) and the overlapping/interchangeable terminology used in patient safety 2. In addition, the inclusion of non-English journals in scientific databases is limited. Moreover, our review included only studies between 2000 and 2013 in order to avoid the replication of the review published in 2002 by Beijer & De Blaey on the same topic 3. Likewise, it was made in order to give a good up-to-date insight into the current situation of patient safety over the last 13 years in relation to new concepts and definitions. This is important especially because definitions/concepts in the patient safety field have a continuous dynamic nature. We do agree with Souza et al. 4 that following all the steps in the Prisma statement and AMSTAR instrument would add more credibility and robustness to systematic reviews. Therefore, we have matched our systematic review against the Prisma statement in relation to title, abstract, introduction, methods, results and discussion. Apart from the risk of bias, the review has been shown to fulfil the Prisma statement. However, we have mentioned this risk in the ‘strengths and limitations of the review section’ and proposed our approach to avoid it. The same scenario applied to the AMSTAR tool for quality assessment; where we fulfilled eight out of 11 items which is comparable with other published reviews in the field. Criteria not fulfilled were bias assessment, list of included/excluded studies and assessing the studies' qualities. Addressing the review replication, Souza et al. 4 did not mention two key words in the title of their letter which were (adult patients). This may have affected the results. By excluding studies carried out on paediatrics initially, we excluded a reasonable number of hits. Therefore, in addition to the two missing key search terms, the reason for the large number of hits found by Souza et al. may be due to the different search strategy being used. In fact, using a too sensitive search strategy can result in many redundant hits and initially increase the researcher's workload. However, our number of hits was consistent more with that of a recent systematic review on drug related hospital admission, made using PubMed by Nivya et al. covering the period between 2007–2013, and they found only (366) initial hits 5. We have some reservations on the comments regarding the causality assessment which were highlighted by Souza et al. 4. Thus, the variation in causality assessment was encountered due to the inevitable variations in definitions/methods across studies. However, we assume that Souza et al. are fully aware of many reviews in the field which included studies with different characteristics and were extremely beneficial to the growing field of patient safety. Furthermore, we preferred the results' segmentation depending on the orientation and the final outcomes of the studies rather than the subgrouping suggested by Souza et al. (study sample, number of medicines or data collection systems) because it presents the findings in a simple and comprehensible way. In addition, it demonstrates the differences across the main studied categories (medicine related problems, adverse drug reactions and adverse drug events) with regards to prevalence, causes and main risk factors involved. Additionally, the meta-analysis did not seem a valid option from our perspective although we tried initially. Souza et al. 4 suggested carrying out meta-analysis for 17 studies out of the 45. However, we have considered meta-analysis as a package which is either to be done for all studies or to be ignored due to the irrelevance 6. Moreover, we have carried out the normality and heterogeneity tests and the data did not show a normal distribution. Therefore, we have used median and IQR instead of mean and confidence interval 7. We would like to thank Souza et al. 4 again for the interest they showed regarding our review.