<p>Final logistic regression models (n = 167).</p>

Background This study aimed to identify the predictive effects of different aspects of diabetic peripheral neuropathy (DPN) and other already known risk factors for falls through a comprehensive logistic model within community-dwelling older adults with diabetes and DPN. This paper also provides a model that estimates the probability of a fall occurring in a real-world clinical scenario. Methods This cross-sectional retrospective study analyzed data from subjects that had never fallen (non-fallers, n = 534) and that had fallen at least twice in the previous year (fallers, n = 101). The logistic regression analysis was performed on a training sample randomly extracted from the original sample (non-fallers: n = 85; fallers: n = 81). The model was validated by checking the performance parameters using a test sample comprised of 10% of fallers (n = 16) and a proportionate subsample of non-fallers (n = 85) from the original dataset. Results Three predictive models were developed. The best model (0.762 receiver operating characteristic[ROC] curve area, 60.4% accuracy, 68.8% sensitivity, 58.8% specificity) identified age (odds ratio[OR]=1.06[95%CI: 1.02, 1.10], P = 0.002), Michigan Neuropathy Screening Instrument score (OR=1.23[95%CI: 1.08, 1.40], P = 0.001), and self-reported balance problems (OR=2.65[95%CI: 1.29, 5.45], P = 0.008) as predictors of falls. A second model with good performance parameters (0.750 ROC curve area, 62.4% accuracy, 62.5% sensitivity, 62.4% specificity) showed that age (OR=1.04[95%CI: 1.01, 1.07], P = 0.015), balance problems (OR=3.29[95%CI: 1.64, 6.59], P = 0.001), and DPN severity (OR=1.18[95%CI: 1.03, 1.34], P = 0.018) were predictors of falls. Conclusions We showed the potential of a predictive model for recurrent falls based on commonly evaluated variables in community-dwelling individuals with diabetes for use in clinical practice. Even for individuals who are not at a high risk for falls, it is crucial to assess the combination of DPN signs, symptoms, and severity and the perception of balance problems, as these are more relevant in people with diabetes than the traditional physical impairments associated to aging.研究背景 本研究旨在针对社区居住的糖尿病合并糖尿病周围神经病变（Diabetic Peripheral Neuropathy，DPN）老年人群，通过综合logistic回归模型，明确DPN不同维度及其他已知跌倒危险因素对跌倒的预测效应。本研究同时构建了可在真实临床场景中估算跌倒发生概率的预测模型。 研究方法 本项横断面回顾性研究分析了两类受试者的数据：从未发生跌倒者（非跌倒组，n=534）以及既往1年内至少跌倒2次者（跌倒组，n=101）。从原始样本中随机抽取训练集（非跌倒组85例，跌倒组81例），并以此开展logistic回归分析。通过测试集验证模型性能：测试集包含原始数据中10%的跌倒受试者（n=16）以及按比例抽取的非跌倒亚组受试者（n=85），并以此计算模型性能参数。 研究结果 本研究共构建3个跌倒预测模型。最优模型的受试者工作特征曲线（Receiver Operating Characteristic，ROC）下面积为0.762，准确率60.4%，灵敏度68.8%，特异度58.8%；该模型筛选出的跌倒预测因子包括年龄（比值比（Odds Ratio，OR）=1.06，95%置信区间（95%CI）：1.02~1.10，P=0.002）、密歇根神经病变筛查量表（Michigan Neuropathy Screening Instrument）评分（OR=1.23，95%CI：1.08~1.40，P=0.001）以及自我报告的平衡障碍（OR=2.65，95%CI：1.29~5.45，P=0.008）。性能次之的模型（ROC曲线下面积0.750，准确率62.4%，灵敏度62.5%，特异度62.4%）显示，年龄（OR=1.04，95%CI：1.01~1.07，P=0.015）、平衡障碍（OR=3.29，95%CI：1.64~6.59，P=0.001）以及DPN严重程度（OR=1.18，95%CI：1.03~1.34，P=0.018）为跌倒预测因子。 研究结论 本研究证实，基于社区居住糖尿病患者常用评估变量构建的复发性跌倒预测模型，具备临床应用潜力。即便对于跌倒高风险以外的人群，评估DPN的体征、症状、严重程度以及平衡障碍感知情况仍至关重要；相较于与衰老相关的传统躯体功能受损，上述指标在糖尿病患者中与跌倒的关联更为显著。