背景:邮政筛查以前尚未被验证为识别社区居民跌倒和骨折风险的方法。我们检查了英国预防跌倒伤害试验(PreFIT;ISRCTN71002650)中使用的邮政风险筛查器的预后表现,预测任何跌倒,反复跌倒,和超过12个月的骨折。我们测试了添加变量是否会提高筛选器性能。
方法:九千八百八万社区居民参与者,70岁及以上,英国国民健康服务(NHS)的63项一般做法被纳入了一项大型的、比较筛查和治疗跌倒预防干预措施的实用群集随机试验。短邮件筛选器作为A4纸发送给试验干预组中的所有参与者,以完成并返回GP(n=6,580)。邮政筛选器项目被嵌入到所有试验组的基线随机化前邮政问卷中(n=9,808)。我们使用曲线下面积(AUC)评估鉴别和校准。我们使用来自控制组的数据确定了其他预测因子,并将这些系数应用于干预组参与者的内部验证模型。我们使用逻辑回归来识别其他预测变量。
结果:在12个月内共报告了10,743例跌倒和307例骨折。超过三分之一的参与者3,349/8,136(41%)在12个月的随访中至少下降了一次。对邮政筛选器的反应很高(5,779/6,580;88%)。预测模型在控制和干预武器中均显示出相似的判别能力,任何下降AUC的判别值0.67(95%CI0.65至0.68),和复发性跌倒(AUC0.71;95%CI0.69,0.72),但对骨折的辨别较差(AUC0.60;95%CI0.56,0.64)。额外的预测变量改善了跌倒的预测,但对骨折的影响不大,其中AUC升至0.71(95%CI0.67至0.74)。校准斜率非常接近1。
结论:在初级保健中使用短期跌倒风险邮政筛查是可以接受的,但跌倒预测有限,虽然与其他工具一致。尽管增加了变量,但骨折和跌倒预测仅部分依赖于跌倒风险。
Postal screening has not previously been validated as a method for identifying fall and fracture risk in community-dwelling populations. We examined prognostic performance of a postal risk screener used in the UK Prevention of Falls Injury Trial (PreFIT; ISRCTN71002650), to predict any fall, recurrent falls, and fractures over 12 months. We tested whether adding variables would improve screener performance.
Nine thousand eight hundred and eight community-dwelling participants, aged 70 years and older, and 63 general practices in the UK National Health Service (NHS) were included in a large, pragmatic cluster randomised trial comparing screen and treat fall prevention interventions. The short postal screener was sent to all participants in the trial intervention arms as an A4 sheet to be completed and returned to the GP (n = 6,580). The postal screener items were embedded in the baseline pre-randomisation postal questionnaire for all arms of the trial (n = 9,808). We assessed discrimination and calibration using area under the curve (AUC). We identified additional predictors using data from the control arm and applied these coefficients to internal validation models in the intervention arm participants. We used logistic regression to identify additional predictor variables.
A total of 10,743 falls and 307 fractures were reported over 12 months. Over one third of participants 3,349/8,136 (41%) fell at least once over 12 month follow up. Response to the postal screener was high (5,779/6,580; 88%). Prediction models showed similar discriminatory ability in both control and intervention arms, with discrimination values for any fall AUC 0.67 (95% CI 0.65 to 0.68), and recurrent falls (AUC 0.71; 95% CI 0.69, 0.72) but poorer discrimination for fractures (AUC 0.60; 95% CI 0.56, 0.64). Additional predictor variables improved prediction of falls but had modest effect on fracture, where AUC rose to 0.71 (95% CI 0.67 to 0.74). Calibration slopes were very close to 1.
A short fall risk postal screener was acceptable for use in primary care but fall prediction was limited, although consistent with other tools. Fracture and fall prediction were only partially reliant on fall risk although were improved with the additional variables.