当前位置: X-MOL 学术Dev. Med. Child Neurol. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Validity of a meta‐analysis of risk factors for neonatal brachial plexus palsies
Developmental Medicine & Child Neurology ( IF 3.8 ) Pub Date : 2020-03-07 , DOI: 10.1111/dmcn.14508
Andreas Rehm 1 , Azeem Thahir 2
Affiliation  

EDITOR–We read with interest the recent systematic review and meta‐analysis by Van der Looven et al. A meta‐ analysis is a quantitative statistical analysis of similar studies which mostly tend to include randomized controlled trials (RCTs), providing the highest validity and the least bias, followed by prospective cohort studies, case–control/retrospective cohort studies, and case series with increasing bias. Multiple strategies and review methodologies are available to search databases and identify studies including the Preferred Reporting Items for Systematic Reviews and Meta‐ Analyses (PRISMA) statement, Meta‐analyses Of Observational Studies in Epidemiology (MOOSE) guidelines, and Grant’s review as used by Van der Looven et al. All have a level of subjectivity since search filters are defined by the researchers, producing varying results for the same database registry search. Van der Looven et al.’s meta‐analysis mixed three different study types: mainly retrospective cohort and case– control studies (3rd tier), and no RCTs. The chosen studies lack similarity regarding reported risk factors, data collection years, and place (four continents, 11 countries), and included hugely varying counts of live births with and without neonatal brachial plexus palsies (NBPP). The database registries collected data for either billing or outcome purposes and varied from single hospital to multiple state and national registries which either used continuous counts or discontinuous extrapolated counts to obtain population estimates. Gutman et al. identified reduced fidelity for a database using data collected for billing and not outcome purposes resulting in underreporting of surgical complications. The huge variation in data collection is reflected in the considerable heterogeneity for the five risk factors of up to 98% and raises doubts about the quality assessment of each study using the Newcastle‐Ottawa Scale, which has been shown to have low reliability between reviewers. Van der Looven et al. reduced the heterogeneity by removing between 17% and 43% of studies from the already decimated number of studies used for the individual risk factor meta‐analyses. This is equivalent to manipulating the eligibility criteria and reduces the validity and increases the bias in interpreting the findings. The challenge in collating studies within systematic reviews which have not used the same methodologies results in difficulties calculating the total number of live births, especially if the studies have utilized the same database. While ideally all studies should have data for all five risk factors, this not being the case here raises issues not only on calculations of total live births but questions how to interpret findings from five separate meta‐analyses as performed by Van der Looven et al. A further issue relates to how population estimates are calculated. Abzug et al. determined findings from extrapolated data from the Kids Inpatient Database (KID) for the 1997 to 2012 period. However the database only yields sample data from hospital bills for 5 564 628 births which are 8.5% of live births born in the United States during the study period. No real data are available for the additional extrapolated 18 594 798 live births. No data were collected for every 2‐year block between the collection years. This limitation needs to be recognized as the missing years of the KID could alter the reported negative trend (https://www.hcup-us.ahrq. gov/db/nation/kid/kid_statelevelestimates.jsp). All included case–control studies used an arbitrary sized control group and random or systematic sampling of their controls without assessing if they were representative for their population. This can lead to unreliable biased results. Abzug et al. reported a decrease of the NBPP rate but the used Swedish registry data showed an increase from 0.17% in 1987 to 0.27% in 1997. All other studies do not provide trends. We do not think that it is valid to conclude that the included studies show that the NBPP rate is decreasing and that the overall incidence was 0.174%, because of the huge variability between the study data which also reduces the reliability of the risk factor data.

中文翻译:

新生儿臂丛神经麻痹危险因素荟萃分析的有效性

编辑——我们饶有兴趣地阅读了 Van der Looven 等人最近的系统评价和荟萃分析。荟萃分析是对类似研究的定量统计分析,主要倾向于包括随机对照试验(RCTs),提供最高的有效性和最小的偏倚,其次是前瞻性队列研究、病例对照/回顾性队列研究和病例系列研究随着偏见的增加。多种策略和评价方法可用于搜索数据库和识别研究,包括系统评价和 Meta 分析的首选报告项目 (PRISMA) 声明、流行病学观察性研究的 Meta 分析 (MOOSE) 指南,以及 Van 使用的格兰特审查der Looven 等人。由于搜索过滤器是由研究人员定义的,所有这些都具有一定程度的主观性,为相同的数据库注册搜索产生不同的结果。Van der Looven 等人的荟萃分析混合了三种不同的研究类型:主要是回顾性队列研究和病例对照研究(第三层),没有 RCT。所选研究在报告的风险因素、数据收集年份和地点(四大洲,11 个国家/地区)方面缺乏相似性,并且包括有和没有新生儿臂丛神经麻痹 (NBPP) 的活产数量差异很大。数据库登记处收集数据用于计费或结果目的,并且从单一医院到多个州和国家登记处不同,这些登记处使用连续计数或不连续外推计数来获得人口估计值。古特曼等人。使用为计费而非结果目的收集的数据发现数据库的保真度降低,导致手术并发症的漏报。数据收集的巨大差异反映在五个风险因素高达 98% 的相当大的异质性上,并引发了对使用纽卡斯尔-渥太华量表进行的每项研究的质量评估的怀疑,该量表已被证明在审稿人之间的可靠性较低。范德卢文等。通过从用于个体风险因素荟萃分析的已经减少的研究数量中删除 17% 到 43% 的研究,减少了异质性。这相当于操纵资格标准,降低了有效性并增加了解释结果的偏差。在未使用相同方法的系统评价中整理研究的挑战导致难以计算活产总数,特别是如果这些研究使用相同的数据库。虽然理想情况下所有研究都应该拥有所有五个风险因素的数据,但这里的情况并非如此,不仅会引起总活产数计算方面的问题,还会产生问题,即如何解释 Van der Looven 等人进行的五项独立荟萃分析的结果。另一个问题涉及如何计算人口估计数。阿布祖格等人。根据 1997 年至 2012 年期间儿童住院数据库 (KID) 的外推数据确定结果。然而,该数据库仅从医院账单中生成 5 564 628 例新生儿的样本数据,即 8 例。5% 的活产婴儿在研究期间在美国出生。没有额外推断的 18 594 798 例活产儿的真实数据。在收集年份之间的每 2 年块中没有收集数据。需要认识到这一限制,因为 KID 的缺失年份可能会改变报告的负面趋势 (https://www.hcup-us.ahrq.gov/db/nation/kid/kid_statelevelestimates.jsp)。所有纳入的病例对照研究均使用任意规模的对照组和对其对照组的随机或系统抽样,而未评估它们是否代表其人群。这可能会导致不可靠的有偏见的结果。阿布祖格等人。报告 NBPP 率下降,但使用的瑞典注册数据显示从 1987 年的 0.17% 增加到 1997 年的 0.27%。所有其他研究都没有提供趋势。
更新日期:2020-03-07
down
wechat
bug