Dynamic Nomogram for Predicting the Fall Risk of Stroke Patients: An Observational Study.

BACKGROUND

Common fall risk assessment scales are not ideal for the prediction of falls in stroke patients. The study aimed to develop and verify a dynamic nomogram model for predicting the falls risk in stroke patients during rehabilitation.

METHODS

An observational study design was adopted, 488 stroke patients were treated in a tertiary hospital from March to September 2022 were investigated for fall risk factors and related functional tests. We followed up by telephone within 2 months after that to understand the occurrence of falls. Forward stepwise regression was used to analyze the data, and a dynamic nomogram model was developed.

RESULTS

During follow-up, three patients died, and 16 failed the follow-up, with a failure rate of 3.89%. Among 469 patients, 115 experienced falls, with a fall incidence rate of 24.4% and a cumulative of 163 falls. The fall risk was higher among patients aged 60-69, and ≥80 years than among patients aged <60 years. Patients with a fall history within the last 3 months, or a Berg balance scale (BBS) score of <40, or combined with anxiety had a higher fall risk. The differentiation of the dynamic nomogram model was evaluated. The area under the receiver operating characteristics curve (AUC-ROC), sensitivity, specificity of the model was 0.756, 66.09% and 73.16%, respectively. The AUC-ROC of the model was 0.761 by using the Bootstrap test, and the calibration curve coincided with the diagonal dashed line with a slope of one. The Hosmer-Lemeshow good of fit test value was ²=2.040, and the decision curve analysis showed that the net benefit was higher than that of the two extreme curves.

CONCLUSION

Independent fall risk factors in stroke patients are age, had a fall history within the last 3 months, anxiety, and with the BBS score below 40 during rehabilitation. The dynamic nomogram prediction model for stroke patients during rehabilitation has good differentiation, calibration, and clinical utility. The prediction model is simple and practical.