Risk prediction models for falls in hospitalized older patients: a systematic review and meta-analysis

Mao, Anli; Su, Jie; Ren, Mingzhu; Chen, Shuying; Zhang, Huafang

doi:10.1186/s12877-025-05688-0

BMC Geriatrics

Table 2 Overview of the information of the included prediction models

From: Risk prediction models for falls in hospitalized older patients: a systematic review and meta-analysis

Author (year)	Continuous variable processing method	Missing data handling	Modelling Methods	Validation method	Model performance			Model presentation	Final predictors	Interpretability
Author (year)	Continuous variable processing method	Missing data handling	Modelling Methods	Validation method	Discrimination	Calibration	Others	Model presentation	Final predictors	Interpretability
Dormosh [16] 2023	Continuous variable	Multiple imputation	LR + Lasso	10-fold cross-validation	AUROC: 0.695 (0.667–0.724)	Calibration curve	-	The formula of risk score	Fall history (0.34), Cardiac arrhythmias (0.35), Renal failure (0.33), Antipsychotics (0.48), Admission to neurologic department (0.53), Admission to emergency department (0.56), Heart rate (0.01), Katz ADL score (0.06), DOS score (0.04), Missing potassium (-0.38), Missing calcium (-0.25), Missing PaCO2 (-0.32), Missing DOS score (-0.41),	-
Adeli [17] 2023	Continuous variable	Imputation	ANN	Leave-one-out cross-validation	AUROC:0.762	-	Acc:0.731; Spe:0.732 Average Precision:0.499 Precision:0.441; Recall:0.728 F1:0.549	Prediction probability	-	NR
Zhao [18]2020	Continuous variable	NR	LR	Random split	AUROC A:0.874(0.784–0.964) B:0.847(0.771–0.924)	Calibration curve	Sen:0.692 Spe:0.855	Nomogram	History of fractures (1.67), Orthostatic hypotension (1.72), Functional status (1.07), Sedative-hypnotics (2.11), Level of serum albumin (2.53)	-
Wijesinghe [19] 2020	NR	NR	LR; SVM; RF	10-fold cross-validation	NR	NR	Precision:0.756; Recall:0.937; F1:0.836	Prediction probability	NR	NR
Kawazoe [20] 2022	Categorical variables	Multiple imputation	BERT + Bi-LSTM	Temporal validation	AUROC:0.851	NR	Sen:0.737; Spe:0.839; Precision:0.093; F1:0.165	Prediction probability	-	NR
Chu [21] 2022	Categorical variables	NR	DNN; XGBoost; LightGBM; RF; SGD; LR	Random split	AUROC:0.694	NR	Acc:0.730; Sens:0.694; Spe:0.694; Precision score:0.694; Recall score:0.694; F1:0.730	Prediction probability	-	PFI
Alharbi [22] 2022	Categorical variables	NR	Catboost	Independent dataset validation	NR	NR	Dataset SERV: Acc:0.942, Sen:0.916, Spe:0.968, PPV:0.968,NPV:0.918,F1:0.941 Dataset SV: Acc:0.989, Sen:0.988, Spe:0.990, PPV:0.992, NPV:0.985, F1:0.990	Software-based platform	-	NR
Peel [23] 2021	Categorical variables	NR	LR	Independent dataset validation	AUROC: 0.700(0.630–0.760)	NR	Sen:0.72 spe:0.60	Scoring table	Sex (0.63), BMI (0.53), Fall in last 90 days (0.51), Balance problem (0.69), Psychological problems (0.86), Age (-0.03)	-
Vratsistas-Curto [24] 2018	Continuous variable	NR	LR	Bootstrap	AUROC: 0.730(0.660–0.810)	NR	-	Scoring table	Mobility/transfers, Mentalstatus/cognition, Male sex.	-
Beauchet [25] 2018	Categorical variables	Exclude	ANN	Random split	NR	NR	Acc:0.838; Sen:0.296; Spe:0.943 PPV:0.500; NPV:0.874; F1:0.372	Prediction probability	-	NR
GholamHosseini [26] 2014	NR	NR	NR	Random split	NR	NR	Acc:0.740; Sen:0.850; PPV:0.850 F1:0.850	Scoring table	Real-time vital signs, Motion data, Medications, Falls history and muscle strength.	NR
Neumann [27] 2013	Continuous variable	-	LR; DT; Add-up model	Temporal validation	NR	NR	Sen:0.460; Spe:0.711; PPV:0.149 NPV:0.923	Scoring table	Mental alteration, Fall history, Insecure mobility	-
Marschollek [28] 2012	Continuous variable	Mean imputation	LR; DT	10-fold cross-validation	AUROC:0.63	NR	Acc:0.660; Sen:0.554; Spe:0.671 PPV:0.150; NPV:0.935; F1:0.237	Prediction probability	High age, Low Barthel index, Cognitive impairment, Multi-medication Co-morbidity	PFI

NR: Not report, Not applicable, A: Training data, B: test data, Multiple imputation: Imputation of missing values was performed multiple times by constructing estimation models, AUROC: Area Under the Receiver Operating Characteristic curve, Nomogram: A graphical interface that calculates the total score of the influencing factors to obtain the final prediction probability, LR: Logistic Regression, DT: Decision Tree, SVM: Support Vector Machine, RF: Random Forest, DNN: Deep neural network, XGBoost: eXtreme Gradient Boosting, LightGBM: Light Gradient Boosting Machine, SGD: Stochastic Gradient Descent, CatBoost: categorical boosting, ANN: Artificial Neural Network. Acc: Accuracy, Sen: Sensitivity, Spe: Specificity, PPV: Positive Predictive Value, NPV: Negative Predictive Value, F1: F1 Score (F1 Score =(2×Precision×Recall)/(Precision + Recall)), BERT: bidirectional encoder representations from transformers, Bi-LSTM: bidirectional long short-term memory, PFI: Permutation Feature Importance

Back to article page

ISSN: 1471-2318

Contact us

General enquiries: journalsubmissions@springernature.com