Skip to main content

In-hospital fall prediction using machine learning algorithms and the Morse fall scale in patients with acute stroke: a nested case-control study



Falls are one of the most common accidents in medical institutions, which can threaten the safety of inpatients and negatively affect their prognosis. Herein, we developed a machine learning (ML) model for fall prediction in patients with acute stroke and compared its accuracy with that of the existing fall risk prediction tool, the Morse Fall Scale (MFS).


This is a retrospective nested case-control study. The initial sample size was 8462 admitted to a single cerebrovascular specialty hospital with acute stroke. A total of 156 fall events occurred, and each fall case was randomly matched with six control cases. Six ML algorithms were used, namely, regularized logistic regression, support vector machine, naïve Bayes (NB), k-nearest neighbors, random forest, and extreme-gradient boosting (XGB).


We included 156 in the fall group and 934 in the non-fall group. The mean ages of the fall and non-fall groups were 68.3 (± 12.2) and 65.3 (± 12.9) years old, respectively. The MFS total score was significantly higher in the fall group (54.3 ± 18.3) than in the non-fall group (37.7 ± 14.7). The area under the receiver operating curve (AUROC) of the MFS in predicting falls was 0.76 (0.73–0.79). XGB had the highest AUROC of 0.85 (0.78–0.92), and XGB and NB had the highest F1 score of 0.44.


The AUROC values of all of ML algorithms were similar to those of the MFS in predicting fall risk in patients with acute stroke, allowing for accurate and efficient fall screening.

Peer Review reports


In-hospital falls are among the most common patient safety incidents in healthcare facilities [1]. They can increase the length of hospital stay, incur additional healthcare costs, and can even lead to legal disputes between the healthcare providers and patients [2]. In their multi-center study, Morello et al. [3] found that in-hospital falls cause an average of eight additional days of stay in hospital and incur an average additional cost of $6669. Furthermore, they negatively affect patients and their families because of increased time and financial burden [4]. The incidence of falls is particularly high among those with cerebrovascular diseases due to impaired postural stability, decreased sensory function, and motor deficits [5]. Previous studies have reported high post-stroke fall rates, with 1.8–14% of patients with stroke experiencing falls during hospitalization [6, 7].

Medical staff must assess fall risk based on a patient’s characteristics to effectively predict their probability [8]. There have been several fall risk assessing tools, such as the St. Thomas Risk Assessment Tool [9], Hendrich II Fall Risk Model [10], Johns Hopkins Fall Risk Assessment Revised Tool [11], and Morse Fall Scale (MFS) [12]. These tools have been developed by determining and categorizing the fall risk factors. However, sufficient staff and time are required to complete these evaluations and they do not sufficiently reflect the characteristics of those patients with potential risks [13, 14]. Among them, the MFS is the most widely utilized tool for assessing the risk of falls in South Korea [13]; it consists of six items, including fall history, secondary diagnosis, use of assistive devices, intravenous or heparin cap, gait, and self-insight related to gait disorders [12, 15]. The MFS has been validated in several studies and is considered a reliable tool for measuring fall risk [16,17,18]. However, it has limitations in predicting fall risk factors in uncooperative patients. Therefore, reflecting the characteristics of a patient’s clinical situation in fall prediction and supplementing the drawbacks or limitations of fall risk assessment tools used in clinical practice are needed to improve fall prediction [19, 20]. Various factors affect the likelihood of falls and, in medical institutions, which treat patients with severe disease, a predictive model for disease-specific fall risk factors is essential. Nevertheless, fall risk screening tools are not sufficient to prevent in-hospital falls [14].

In recent years, there has been a rapid increase in medical research based on machine learning (ML) [21]. It is primarily used for implementing prediction models; however, its scope is expanding to include the classification of disease severity [22], medical decision-making [23], and application of newly developed therapeutic interventions [24]. An advantage of ML-based models is their ability to predict a patient’s prognosis or progress in a specific situation based on data from the electronic health records (EHRs) [25]. ML is able to integrate clinical information in a meaningful manner, providing medical staff with comprehensive information for ensuring fully informed medical decision-making [26]. Previous studies have shown that ML-based algorithms can produce results equivalent to, or better than, those produced by traditional tools if sufficient data and appropriate algorithms are used [27, 28]. However, to the best of our knowledge, there have been no studies till date presenting an ML-based fall prediction model in hospitalized patients with acute stroke.

This study aimed for the following: (1) to develop an ML model for prediction of in-hospital fall risk among patients with acute stroke; (2) to compare the ML model’s predictive performance with that of the existing fall risk assessment tool, the MFS.


Data source and patient inclusion

This retrospective study utilized EHRs to identify patients who were admitted to a single cerebrovascular specialty hospital between January 2016 and June 2022 with a primary diagnosis of acute stroke, as defined by the International Classification of Diseases-10 codes I60–I63. We initially identified 8462 patients. During this period, 156 fall events occurred (1.84%). If a significant difference in frequency was found between the fall event group and the control group, a retrospective nested case-control study was performed using random sampling methods, which were frequently used in previous related studies [20, 29]. Each fall case was randomly matched with six control groups (n = 936), with matching performed based on admission in the same quarter and ward. Cases with missing values were excluded from the study (Fig. 1). For the robustness of the statistical analysis, if there were two or more fall events in one admission, the first fall was used as the index.

Fig. 1
figure 1

Flowchart of patient inclusion

This study design was reviewed and approved by the Institutional Review Board of Pohang Stroke and Spine Hospital (Approval No. PSSH0475-202108-HR-016-04). The informed consent was waived by the Institutional Review Board of Pohang Stroke and Spine Hospital due to the retrospective nature of this study and anonymity of the database. The study was conducted following the principles of the Declaration of Helsinki.

Study variables

Evaluation indicators assessed during the initial hospitalization were used as the main variables to predict falls. Age and sex were identified as basic information. Body mass index (BMI), haemoglobin level, and albumin level were checked to reflect the patient’s nutritional status. Stroke subtypes were classified as subarachnoid hemorrhage (I60), intracerebral hemorrhage (I61 and I62), or ischemic stroke (I63). Finally, the National Institutes of Health Stroke Scale (NIHSS) score was assessed to identify stroke severity.

As factors reflecting the patient’s status at admission, the differences in admission route, admission method, and ward type were assessed. As a result, the admission route was divided into emergency room and outpatient admission, and the admission method was classified into walking, wheelchair, and bedridden. In addition, the initial admission ward was classified into general ward, integrated nursing care service (INCS), and special care units – intensive care unit (ICU) and stroke unit (SU).

Socioeconomic factors were divided into the medical insurance type and residence area. The medical insurance type was classified into medical aid and national health insurance coverage. According to the Korean administrative distinct, the residential area was divided into the “dong” and “eup/myeon.” Accompanying diseases such as hypertension, diabetes, dyslipidemia, arrhythmia, cardiovascular diseases, osteoporosis, degenerative spinal disease, and neurodegenerative brain disease were assessed (Table S1). The prescribed drugs were checked with the standard drug code name and the Anatomical Therapeutic Chemical Classification System developed by the World Health Organization. In the fall group, drugs administered on the day of the fall event were included, and in the control group, drugs administered at the time of admission were included. Medications were categorized into antidepressants, anxiolytics, antipsychotics, antiepileptic drugs, and diuretics, and patients were classified as those without medication history in the category, those taking only one type of medication, and those taking multiple classes of medications.

Finally, our ML models were compared with the existing fall risk prediction tool, the MFS, which was evaluated by skilled nursing staff at the time of patient admission. The total MFS scores, a routine assessment of fall screening in the setting of this study, was used to predict falls. The list of all variables used in the predictive model is summarized in Table S2.

Statistical analysis

This study used the R software version 4.3.0 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) for all statistical and ML analyses. Continuous variables were presented as mean ± standard deviation, and categorical variables were presented as frequencies (percentages). For comparison between the fall and non-fall groups, independent t-tests were performed for continuous variables, and chi-square (trend) tests were performed for categorical variables. P-values of < 0.05 were considered statistically significant. Univariable binary logistic regression models were applied to evaluate the predictive power of the MFS, and the area under the receiver operating characteristic curve (AUROC) was calculated and compared with other ML models.

To investigate the relationship between fall occurrence and variables, a binary logistic regression model was established. Variable selection was performed using stepwise backward elimination, and the Akaike information criterion was used as an estimator of multivariable model fitness. The measurement of multicollinearity was conducted using the criterion of sqrt (variable inflation factor) > 2.

Data pre-processing and ML process

Data pre-processing was performed for the ML prediction model. First, variables with low frequency and those showing multicollinearity were identified. For continuous variables, centering and scaling were performed. One-hot encoding was applied to convert categorical variables into numeric variables. Data were randomly divided into training and validation data at a 2:1 ratio. To balance the dependent variable, training data were oversampled using the synthetic minority oversampling technique. Six ML algorithms were used for the ML process, namely, regularized logistic regression (RLR), support vector machine (SVM), naïve Bayes (NB), k-nearest neighbors (KNN), random forest (RF), and extreme-gradient boosting (XGB). For internal validation, 10-fold cross-validation was repeated 50 times using training data. Hyperparameter tuning was conducted using a combination of random and grid searches (Table S3). To assess their prediction performance in terms of AUROC, F1 score, sensitivity, specificity, positive predictive value, and negative predictive value, the optimal trained models for each algorithm were applied to the validation data. Finally, feature importance was measured for the RLR, RF, and XGB models (Fig. 2). The “caret” package in R software was used for the ML process [30]. The entire code for this study is provided in the online supplementary materials.

Fig. 2
figure 2

Frame of machine learning prediction for this study


Baseline characteristics and the Morse fall scale

In our final analysis, there were 156 and 934 patients in the fall and non-fall groups, respectively. Table 1 summarizes the baseline characteristics of the patients in the fall and non-fall groups. The features of the in-hospital falls recorded are summarized in Table 2.

Table 1 Baseline characteristics
Table 2 Features of in-hospital falls in this study

The mean MFS score was significantly higher in the fall group (54.3 ± 18.3) than in the non-fall group (37.7 ± 14.7). The AUROC of the MFS in predicting falls was 0.76 (0.73–0.79), and the sensitivity and specificity were 0.72 (0.65–0.79) and 0.74 (0.71–0.77), respectively. The cutoff value for predicting falls using the mean MFS score was 42.50 points.

Stepwise logistic regression model

Table 3 presents the final binary logistic regression model after stepwise backward elimination. Figure 3 shows the distribution of adjusted odds ratios (aOR) and 95% confidence intervals (CI) for each variable. The type of ward was significantly associated with a lower risk of falls in the INCS, whereas the ICU/SU was associated with a higher risk of falls. In addition, admission with wheelchair ambulation, diabetes, arrhythmia, degenerative spinal diseases, cerebral neurodegenerative diseases, and medications were significantly associated with a higher risk of falls. In comparison, dyslipidemia and alert mental status were significantly associated with a lower risk of falls.

Table 3 Final logistic regression model after stepwise backward elimination
Fig. 3
figure 3

Forest plot of the final stepwise logistic regression model for predicting in-hospital falls

ML prediction

Variables with zero variance, such as osteoporosis, cardiovascular disease, and degenerative spinal diseases, were excluded from the analysis. No evidence of multicollinearity was noted among the continuous variables with correlation coefficients of ≥ 0.7. The ratios of falls to non-falls in the training and validation datasets were 104:626 and 52:308, respectively. After applying the synthetic minority oversampling technique to the training dataset, the revised ratio of falls to non-falls became 624:626. Table S4 presents the confusion matrix for all prediction models.

Among the six ML algorithms, XGB had the highest AUROC of 0.85 (0.78–0.92), and XGB and NB had the highest F1 score of 0.44. The KNN-based prediction model had the highest sensitivity of 0.71 (0.58–0.82), whereas XGB had the highest sensitivity at 0.65 (0.46–0.81). RF showed the highest positive predictive value of 0.85 (0.58–0.96), and KNN showed the highest negative predictive value of 0.94 (0.90–0.96). All ML algorithms showed similar or slightly improved AUROC values compared with MFS (Table 4).

Table 4 Machine learning prediction performance

The NIHSS was the most important feature in predicting falls in all models, including RLR, RF, and XGB. Other variables such as age, BMI, albumin, and hemoglobin were also important predictors. Ward type was a significant variable for predicting falls. In addition, medications and arrhythmia were identified as the top five variables in the RLR model (Fig. 4).

Fig. 4
figure 4

Feature importance analysis revealing the top five most important variables in regularized logistic regression (RLR), random forest (RF), and extreme gradient boosting (XGB) algorithms. Stroke severity was the most important for predicting in-hospital falls


This study proposed ML-based models for predicting in-hospital falls in acute stroke using EHRs. The models demonstrated comparable performance to the MFS in predicting falls. Previous studies using ML to predict falls in hospitalized patients have reported valid results [31,32,33]. Wang et al. [34] reported a robust fall prediction with multi-view ensemble learning with missing values, and their model showed an AUROC of 0.81, which was similar to ours. Nakatani et al. [29] presented a natural language process-based inpatient fall prediction model using EHRs and reported an AUROC of 0.84, which was similar to ours. Our results show that disease-specific variables are essential predictors of falls in this patient group and can improve the accuracy of fall prediction. Furthermore, our findings suggest that ML algorithms can be tailored to specific healthcare settings and disease populations to develop more accurate prediction models. Such prediction models may be critical in reducing fall-related injuries and, ultimately, improving patient outcomes.

Moreover, developing and applying a fall prediction model using ML algorithms has clinical significance in improving the efficiency of medical staff. Nursing staff feel much stress and limitations when assessing and intervening for fall risk with assessment tools [35]. Furthermore, identifying fall risk factors based on the characteristics of each patient requires time and can become an excessive burden [36]. In actual clinical practice, it is difficult for nursing staff to search and find individual risk factors for falls for each patient and provide nursing care accordingly. To overcome these limitations, the use of ML algorithms to predict falls provides an easy and fast way to obtain accurate results. Therefore, this approach has significant clinical significance, enabling nursing staff to predict falls quickly and accurately and intervene accordingly, reducing fall occurrences.

One notable finding among the critical risk factors for falls in patients with acute stroke was the ward type, which was particularly important in INCS. Previous studies in South Korea have yielded inconsistent results regarding the relationship between fall rates and INCS, with some showing higher rates and others showing no significant difference [37, 38]. The present study proved that INCS significantly reduced the risk of falls in patients hospitalized with acute stroke. Thus, the characteristics of patients with acute stroke, most of whom show varying degrees of neurological impairment, may have contributed to these results. In cerebrovascular specialty hospitals, INCS might have focused on fall prevention activities on such disease characteristics. However, more studies are needed to explore this relationship further.

BMI can reflect nutritional status [39], and our results that BMI was one of the critical variables to predict in-hospital falls in patients with acute stroke can indicate that falls may occur frequently in patients with low body weight or weakened physical motor function [40]. Among our results, albumin and hemoglobin levels were found to be important variables for fall risk. Previous studies have reported low albumin levels and anemia as risk factors for falls in patients hospitalized in the acute phase, and these could be equally applied to patients with acute stroke. Finally, socioeconomic status, a well-known risk factor, was found to be unrelated to the in-hospital falls in this study [41, 42]. These results were attributed to the reason that this study was conducted in a single region and incorporated only patients with acute stroke. Therefore, we consider that disease characteristics may make a greater contribution to the risk of falls than socioeconomic characteristics.

Medication use, a well-known fall predictor, was another critical variable in our analysis. Previous studies have shown that medication use, including analgesics, sedatives, vasodilators, and muscle relaxants, is a significant risk factor for falls [43, 44]. Further, polypharmacy increases fall risk [45]. This is particularly relevant for patients with acute stroke because they often have comorbid conditions and receive multiple medications, including central nervous system medications, sedatives, and narcotics, all associated with increased fall risk [46].

In the present study, ensemble models – RF and XGB – showed slightly higher AUROC values but generally lower sensitivity. Conversely, more classical ML algorithms such as the RLR and KNN showed decent AUROC values, along with balanced sensitivity and specificity. This can be attributed to the regularization and relatively simple classification methods overcoming overfitting better than the tree-based ensemble models in this dataset [47]. However, these results cannot be generalized, and more studies based on various databases are needed. Further, this model is intended for screening to prevent falls and is very cost-effective. However, the cost can be much greater once a fall event occurs. Therefore, even if the sensitivity is relatively low, their high specificity and negative predictive value can provide clues for nursing staff to select and focus on patients who need to focus more on fall-prevention activities during their hospitalization [48].

This study is the first to develop an ML-based fall prediction model for patients with acute stroke. We were able to present validated results of ML prediction by comparing them with the MFS, which is the most widely utilized existing fall prediction tool. Furthermore, using multiple ML algorithms for prediction made it possible to directly compare each model’s performance.

This study has several limitations. First, this was a single-center study, which may have limited generalizability. More studies using big data from multiple institutions are needed to verify the results and improve generalizability. Second, this retrospective study used EHRs, which might result in ambiguity in defining some variables. Third, the dataset only observed falls during hospitalization for acute stroke and did not provide long-term follow-up outcomes. Fourth, the timing drug information collection was different between groups. That is, in the fall group, when an event occurred, the medication list was identified with the index date, but in the non-fall group, it was identified with the admission date as the index date. This may have been a source of bias. Finally, despite various statistical adjustments, the outcome variable, in-hospital falls, has a highly imbalanced ground truth, making it difficult to establish causality.


In this study, the ML algorithms used for predicting in-hospital falls among patients with acute stroke showed valid results. Their prediction performance was not equivalent to that of the MFS and they can be readily applied and overcoming the disadvantages of the MFS at the same time. Furthermore, the ML models integrate initial clinical information in a meaningful direction to enable the construction of prediction models that can be used at the beginning of hospitalization. Therefore, the use of ML models for fall prediction is of great clinical significance in allowing medical staff to perform more accurate and efficient fall screening. Ultimately, this study provided cornerstone data for the practical use of the fall screening model of patients with acute stroke in real clinical settings base.

Data Availability

The dataset and entire code supporting the conclusions of this article are included within this article and its additional files.



adjusted odds ratio


area under the receiver operating characteristic curve


body mass index


confidence intervals


electronic health records


intensive care unit


integrated nursing care service


k-nearest neighbors


Morse fall scale


machine learning


naïve Bayes


random forest


regularized logistic regression


stroke unit


support vector machine


extreme-gradient boosting


  1. Schoberer D, Breimaier HE, Zuschnegg J, Findling T, Schaffer S, Archan T. Fall prevention in hospitals and nursing homes: clinical practice guideline. Worldviews Evid Based Nurs. 2022;19:86–93.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Peel NM. Epidemiology of falls in older age. Can J Aging. 2011;30:7–19.

    Article  PubMed  Google Scholar 

  3. Morello RT, Barker AL, Watts JJ, Haines T, Zavarsek SS, Hill KD, et al. The extra resource burden of in-hospital falls: a cost of falls study. Med J Aust. 2015;203:367.

    Article  PubMed  Google Scholar 

  4. Shiffman J. A social explanation for the rise and fall of global health issues. Bull World Health Organ. 2009;87:608–13.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Quigley PA. Redesigned fall and injury management of patients with Stroke. Stroke. 2016;47:e92–4.

    Article  PubMed  Google Scholar 

  6. Nyberg L, Gustafson Y. Fall prediction index for patients in Stroke rehabilitation. Stroke. 1997;28:716–21.

    Article  CAS  PubMed  Google Scholar 

  7. Tutuarima JA, van der Meulen JH, de Haan RJ, van Straten A, Limburg M. Risk factors for falls of hospitalized Stroke patients. Stroke. 1997;28:297–301.

    Article  CAS  PubMed  Google Scholar 

  8. Joint Commission. Preventing falls and fall-related injuries in health care facilities. Sentin Event Alert. 2015;55:1–5.

    Google Scholar 

  9. Oliver D, Britton M, Seed P, Martin FC, Hopper AH. Development and evaluation of evidence based risk assessment tool (STRATIFY) to predict which elderly inpatients will fall: case-control and cohort studies. BMJ. 1997;315:1049–53.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Hendrich A, Nyhuis A, Kippenbrock T, Soja ME. Hospital falls: development of a predictive model for clinical practice. Appl Nurs Res. 1995;8:129–39.

    Article  CAS  PubMed  Google Scholar 

  11. Poe SS, Cvach M, Dawson PB, Straus H, Hill EE. The johns hopkins fall risk assessment tool: postimplementation evaluation. J Nurs Care Qual. 2007;22:293–8.

    Article  PubMed  Google Scholar 

  12. Jewell VD, Capistran K, Flecky K, Qi Y, Fellman S. Prediction of falls in acute care using the Morse fall risk scale. Occup Ther Health Care. 2020;34:307–19.

    Article  PubMed  Google Scholar 

  13. Choi EH, Ko MS, Yoo CS, Kim MK. Characteristics of fall events and fall risk factors among inpatients in general hospitals in Korea. J Korean Clin Nurs Res. 2017;23:350–60.

    Google Scholar 

  14. Morris R, O’Riordan S. Prevention of falls in hospital. Clin Med (Lond). 2017;17:360–2.

    Article  PubMed  Google Scholar 

  15. Baek S, Piao J, Jin Y, Lee SM. Validity of the Morse Fall Scale implemented in an electronic medical record system. J Clin Nurs. 2014;23:2434–40.

    Article  PubMed  Google Scholar 

  16. Chow SK, Lai CK, Wong TK, Suen LK, Kong SK, Chan CK, et al. Evaluation of the Morse fall scale: applicability in Chinese hospital populations. Int J Nurs Stud. 2007;44:556–65.

    Article  PubMed  Google Scholar 

  17. Kim KS, Kim JA, Choi YK, Kim YJ, Park MH, Kim HY, et al. A comparative study on the validity of fall risk assessment scales in Korean hospitals. Asian Nurs Res. 2011;5:28–37.

    Article  Google Scholar 

  18. Urbanetto JS, Pasa TS, Bittencout HR, Franz F, Rosa VP, Magnago TS. Analysis of risk prediction capability and validity of Morse fall scale Brazilian version. Rev Gaucha Enferm Brazilian Version. 2017;37:e62200.

    Google Scholar 

  19. Olsson E, Löfgren B, Gustafson Y, Nyberg L. Validation of a fall risk index in Stroke rehabilitation. J Stroke Cerebrovasc Dis. 2005;14:23–8.

    Article  CAS  PubMed  Google Scholar 

  20. Najafpour Z, Godarzi Z, Arab M, Yaseri M. Risk factors for falls in hospital in-patients: a prospective nested case control study. Int J Health Policy Manag. 2019;8:300–6.

    Article  PubMed  PubMed Central  Google Scholar 

  21. Weissler EH, Naumann T, Andersson T, Ranganath R, Elemento O, Luo Y, et al. The role of machine learning in clinical research: transforming the future of evidence generation. Trials. 2021;22:537.

    Article  PubMed  PubMed Central  Google Scholar 

  22. Park D, Kim BH, Lee SE, Kim DY, Kim M, Kwon HD, et al. Machine learning-based approach for Disease severity classification of carpal tunnel syndrome. Sci Rep. 2021;11:17464.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  23. Sanchez-Martinez S, Camara O, Piella G, Cikes M, González-Ballester MÁ, Miron M, et al. Machine learning for clinical decision-making: challenges and opportunities in cardiovascular imaging. Front Cardiovasc Med. 2021;8:765693.

    Article  PubMed  Google Scholar 

  24. Lewanowicz A, Wiśniewski M, Oronowicz-Jaśkowiak W. The use of machine learning to support the therapeutic process - strengths and weaknesses. Postep Psychiatr Neurol. 2022;31:167–73.

    PubMed  Google Scholar 

  25. Adkins DE. Machine learning and electronic health records: a paradigm shift. Am J Psychiatry. 2017;174:93–4.

    Article  PubMed  PubMed Central  Google Scholar 

  26. Cipriano LE. Evaluating the impact and potential impact of machine learning on medical decision making. Med Decis Making. 2023;43:147–9.

    Article  PubMed  Google Scholar 

  27. Park JH. Machine-learning algorithms based on screening tests for mild cognitive impairment. Am J Alzheimers Dis Other Demen. 2020;35:1533317520927163.

    Article  PubMed  Google Scholar 

  28. Park D, Jeong E, Kim H, Pyun HW, Kim H, Choi YJ et al. Machine learning-based three-month outcome prediction in acute ischemic Stroke: a single cerebrovascular-specialty hospital study in South Korea. Diagnostics (Basel) 2021;11.

  29. Nakatani H, Nakao M, Uchiyama H, Toyoshiba H, Ochiai C. Predicting inpatient falls using natural language processing of nursing records obtained from Japanese electronic medical records: case-control study. JMIR Med Inform. 2020;8:e16970.

    Article  PubMed  PubMed Central  Google Scholar 

  30. Kuhn M, CaReT. Classification and regression training. R package version 6. 0–90. 2021.

  31. Lindberg DS, Prosperi M, Bjarnadottir RI, Thomas J, Crane M, Chen Z, et al. Identification of important factors in an inpatient fall risk prediction model to improve the quality of care using EHR and electronic administrative data: a machine-learning approach. Int J Med Inform. 2020;143:104272.

    Article  PubMed  PubMed Central  Google Scholar 

  32. Patterson BW, Engstrom CJ, Sah V, Smith MA, Mendonça EA, Pulia MS, et al. Training and interpreting machine learning algorithms to evaluate fall risk after emergency department visits. Med Care. 2019;57:560–6.

    Article  PubMed  PubMed Central  Google Scholar 

  33. Thapa R, Garikipati A, Shokouhi S, Hurtado M, Barnes G, Hoffman J, et al. Predicting falls in long-term care facilities: machine learning study. JMIR Aging. 2022;5:e35373.

    Article  PubMed  PubMed Central  Google Scholar 

  34. Wang L, Xue Z, Ezeana CF, Puppala M, Chen S, Danforth RL, et al. Preventing inpatient falls with injuries using integrative machine learning prediction: a cohort study. NPJ Digit Med. 2019;2:127.

    Article  PubMed  PubMed Central  Google Scholar 

  35. Brians LK, Alexander K, Grota P, Chen RW, Dumas V. The development of the RISK tool for fall prevention. Rehabil Nurs. 1991;16:67–9.

    Article  CAS  PubMed  Google Scholar 

  36. Dubbeldam R, Lee YY, Pennone J, Mochizuki L, Le Mouel C. Systematic review of candidate prognostic factors for falling in older adults identified from motion analysis of challenging walking tasks. Eur Rev Aging Phys Act. 2023;20:2.

    Article  PubMed  PubMed Central  Google Scholar 

  37. Yoon S-J, Lee C-K, Jin I-S, Kang J-G. Incidence of falls and risk factors of falls in inpatients. Qual Improv Health Care. 2018;24:2–14.

    Article  Google Scholar 

  38. Jung YA, Sung KM. A comparison of patients’ nursing service satisfaction, hospital commitment and revisit intention between general care unit and comprehensive nursing care unit. J Korean Acad Nurs Adm 2018;24.

  39. Bechard LJ, Duggan C, Touger-Decker R, Parrott JS, Rothpletz-Puglia P, Byham-Gray L, et al. Nutritional status based on body mass index is associated with morbidity and mortality in mechanically ventilated critically ill children in the PICU. Crit Care Med. 2016;44:1530–7.

    Article  PubMed  PubMed Central  Google Scholar 

  40. Yi SW, Kim YM, Won YJ, Kim SK, Kim SH. Association between body mass index and the risk of falls: a nationwide population-based study. Osteoporos Int. 2021;32:1071–8.

    Article  PubMed  Google Scholar 

  41. Kim T, Choi SD, Xiong S. Epidemiology of fall and its socioeconomic risk factors in community-dwelling Korean elderly. PLoS ONE. 2020;15:e0234787.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Mikos M, Trybulska A, Czerw A. Falls – The socio-economic and medical aspects important for developing prevention and treatment strategies. Ann Agric Environ Med. 2021;28:391–6.

    Article  PubMed  Google Scholar 

  43. Michalcova J, Vasut K, Airaksinen M, Bielakova K. Inclusion of medication-related fall risk in fall risk assessment tool in geriatric care units. BMC Geriatr. 2020;20:454.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  44. de Jong MR, Van der Elst M, Hartholt KA. Drug-related falls in older patients: implicated Drugs, consequences, and possible prevention strategies. Ther Adv Drug Saf. 2013;4:147–54.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Ie K, Chou E, Boyce RD, Albert SM. Fall risk-increasing Drugs, polypharmacy, and falls among low-income community-dwelling older adults. Innov Aging. 2021;5:igab001.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Abdollahi M, Whitton N, Zand R, Dombovy M, Parnianpour M, Khalaf K, et al. A systematic review of fall risk factors in Stroke survivors: towards improved assessment platforms and protocols. Front Bioeng Biotechnol. 2022;10:910698.

    Article  PubMed  PubMed Central  Google Scholar 

  47. Subramanian J, Simon R. Overfitting in prediction models - is it a problem only in high dimensions? Contemp Clin Trials. 2013;36:636–41.

    Article  PubMed  Google Scholar 

  48. Trevethan R. Sensitivity, specificity, and predictive values: foundations, pliabilities, and pitfalls in research and practice. Front Public Health. 2017;5:307.

    Article  PubMed  PubMed Central  Google Scholar 

Download references


The authors would like to thank all nurses in Pohang Stroke and Spine Hospital for their dedication and support for patients as well as this study.



Author information

Authors and Affiliations



JHC and DP conceptualized the study. DP contributed to the methodology as well. ESC validated the results and DP performed the formal analysis and contributed in data curation. JHC contributed in investigation and data curation. JHC and DP contributed in writing - original draft. ESC and DP contributed to writing - review and editing. JHC and DP also contributed to visualization, and ESC did the supervision. All authors have read and approved the final manuscript.

Corresponding authors

Correspondence to Eun Suk Choi or Dougho Park.

Ethics declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study design was reviewed and approved by the Institutional Review Board (IRB) of Pohang Stroke and Spine Hospital (approval no. PSSH0475-202108-HR-016-04). The informed consent requirement was waived by the IRB owing to the retrospective nature of this study and anonymity of the database. The study was conducted following the principles of the Declaration of Helsinki.

Consent for publication

Not applicable.

Sponsor’s role

None reported.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Choi, J.H., Choi, E.S. & Park, D. In-hospital fall prediction using machine learning algorithms and the Morse fall scale in patients with acute stroke: a nested case-control study. BMC Med Inform Decis Mak 23, 246 (2023).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: