Dataset - Comparative Effectiveness of Surgical and Combined Interventions for Mandibular Angle Fractures: A Frequentist Network Meta-Analysis

This dataset was curated to support CINeMA assessments and complementary network meta-analysis for two subnetworks (“M” and “S”). Data were provided in two CSV files (cinema_M.csv, cinema_S.csv) at the study–arm level, including study and arm identifiers, treatment codes, numbers of responders and total participants, a risk-of-bias code (1 = low, 2 = some concerns, 3 = high), and an indicator of indirectness when available.

For the M subnetwork, evidence was assembled from 7 studies and 44 arms across 5 treatments, comprising 647 participants and 639 responders. Arm-level risk of bias was distributed as 13.6% low, 72.7% some concerns, and 13.6% high. The most frequently observed direct comparison was M1S versus M3D, followed by M1S versus M2 and M1L versus M1S. For the S subnetwork, 13 studies and 63 arms across 7 treatments were included, totaling 2,107 participants and 1,986 responders. The risk-of-bias profile indicated 71.4% low and 28.6% some concerns, with no high-risk arms. The most common direct comparison involved S1S versus S2, with additional emphasis on S1L versus S1S and S1L versus S3D.

Beyond risk-of-bias profiling, the network geometry was characterized to describe the distribution of evidence across treatments and comparisons, revealing centrally connected interventions with comparatively larger information contributions and peripheral nodes with sparser evidence. Random-effects network meta-analysis models were fitted to estimate pooled relative effects, and between-study heterogeneity was quantified to reflect variability beyond sampling error. Consistency was examined using global and local approaches (including design-by-treatment evaluations and node-splitting where appropriate) to assess agreement between direct and indirect evidence. Small-study effects were explored with comparison-adjusted and network funnel plots. Treatment ranking probabilities were estimated and summarized using SUCRA to communicate the relative standing of interventions under model assumptions. Sensitivity analyses were conducted to assess robustness, including the exclusion of high risk-of-bias studies, alternative assumptions for heterogeneity, and restriction to adequately informed contrasts.

Overall, the evidence base in both subnetworks was predominantly characterized by low to moderate risk of bias, with the principal methodological concern concentrated in the M subnetwork for the M1S versus M3D contrast due to the presence of high and moderate risk contributions. The S subnetwork displayed a more favorable profile without high-risk arms. The dataset enabled reproducible derivation of network structures, risk-of-bias summaries by treatment and comparison, model-based effect estimates, consistency diagnostics, ranking summaries, and robustness checks, supported by accompanying figures that documented network structure and risk-of-bias distributions.

本数据集专为支持CINeMA评估以及针对两个子网络（“M”与“S”）的配套网络荟萃分析而构建。数据以两个逗号分隔值（CSV）文件（cinema_M.csv、cinema_S.csv）的形式提供，粒度为研究-分组水平，包含研究与分组标识符、治疗代码、应答者人数与总参与者数、偏倚风险评分（1=低风险，2=存在一定担忧，3=高风险），以及可用情况下的间接性指标。

对于M子网络，证据汇集自5种治疗方案下的7项研究、44个分组，共纳入647名参与者与639名应答者。分组水平的偏倚风险分布为：13.6%为低风险，72.7%存在一定担忧，13.6%为高风险。最常见的直接对比为M1S与M3D，其次为M1S与M2以及M1L与M1S。对于S子网络，纳入7种治疗方案下的13项研究、63个分组，总计2107名参与者与1986名应答者。其偏倚风险特征为：71.4%为低风险，28.6%存在一定担忧，无高风险分组。最常见的直接对比为S1S与S2，此外重点关注的对比包括S1L与S1S以及S1L与S3D。

除偏倚风险特征分析外，本研究还对网络几何结构进行了表征，以描述证据在各治疗方案与对比间的分布情况，结果显示核心连接的干预措施具有相对更大的信息贡献，而外围节点的证据则较为稀疏。我们构建了随机效应网络荟萃分析模型以合并估计相对效应，并量化了研究间异质性以反映抽样误差之外的变异。采用全局与局部方法检验一致性（包括设计-治疗评估及适用情况下的节点拆分法），以评估直接与间接证据的一致性。通过对比校正漏斗图与网络漏斗图探索小样本效应。基于模型假设，我们估计并通过表面下累积排序概率（SUCRA，Surface Under the Cumulative Ranking Area）总结了治疗排序概率，以直观呈现各干预措施的相对排名。我们还开展了敏感性分析以评估结果的稳健性，包括剔除高偏倚风险研究、采用异质性替代假设，以及限定仅纳入信息充分的对比。

总体而言，两个子网络的证据基础整体以低至中度偏倚风险为主要特征，其中M子网络中M1S与M3D的对比存在高与中度偏倚风险贡献，成为主要的方法学关注点。S子网络则表现出更优的偏倚风险特征，无高风险分组。本数据集支持可复现地推导网络结构、按治疗与对比分组的偏倚风险汇总、基于模型的效应估计、一致性诊断、排序结果汇总以及稳健性检验，配套的可视化图表可辅助展示网络结构与偏倚风险分布。