Development of an assessment sheet for fall prediction in stroke inpatients in convalescent rehabilitation wards in Japan

Abstract

Objective We conducted a study to develop an assessment sheet for fall prediction in stroke inpatients that is handy and reliable to help ward staff to devise a fall prevention strategy for each inpatient immediately upon admission. Methods The study consisted of three steps: (1) developing a data sampling form to record variables related to risk of falls in stroke inpatients and conducting a follow-up survey for stroke inpatients from their admission to discharge by using the form; (2) carrying out analyses of characteristics of the present subjects and selecting variables showing a high hazard ratio (HR) for falls using the Cox regression analysis; (3) developing an assessment sheet for fall prediction involving variables giving the integral coefficient for each variable in accordance with the HR determined in the second step. Results and discussion (1) Subjects of the present survey were 704 inpatients from 17 hospitals including 270 fallers.(2) We selected seven variables as predictors of the first fall: central paralysis, history of previous falls, use of psychotropic medicines, visual impairment, urinary incontinence, mode of locomotion and cognitive impairment. (3) We made 960 trial models in combination with possible coefficients for each variable, and among them we finally selected the most suitable model giving coefficient number 1 to each variable except mode of locomotion, which was given 1 or 2. The area under the ROC curve of the selected model was 0.73, and sensitivity and specificity were 0.70 and 0.69, respectively (4/5 at the cut-off point). Scores calculated from the assessment sheets of the present subjects by adding coefficients of each variable showed normal distribution and a significantly higher mean score in fallers (4.94 ± 1.29) than in non-fallers (3.65 ± 1.58) (P = 0.001). The value of the Barthel Index as the index of ADL of each subject was indicated to be in proportion to the assessment score of each subject. Conclusion We developed an assessment sheet for fall prediction in stroke inpatients that was shown to be available and valid to screen inpatients with risk of falls immediately upon admission.

Systematic review of risk prediction models for falls after stroke

“…The 12 included articles presented 18 risk prediction models. The geographic distribution of the studies was: Europe,12 13 22 23 USA,24 25 27 28 Australasia3 20 and Asia 21 26. Sample sizes ranged from 3224 to 110420 participants, with a total of 4315 participants across all studies.…”

“…Sample sizes ranged from 3224 to 110420 participants, with a total of 4315 participants across all studies. Eight of the models were developed to predict the occurrence of any fall,13 20 21 24–26 while the remaining models focused on multiple or injurious falls, time taken to fall or number of falls 3 12 20 22–24 27 28. Seven studies derived nine models in total to predict falls occurring in the inpatient setting after stroke, with the patients in most studies being followed up for the duration of their hospital stay 21–26 28.…”

“…Discrimination was summarised with the C-statistic (area under the curve, AUC) for seven models in four studies 12 13 20 21. Values ranged from 0.6220 to 0.87 13.…”

“…This was defined by some in terms of time since stroke,20 22 23 25–27 or alternatively by a service transition including rehabilitation admission13 21 24 or discharge 3 12. In order to minimise heterogeneity, studies were excluded where the majority of participants were measured after 3 months since stroke.…”

Aktivnosti za preprečevanje padcev v rehabilitaciji pacientov po možganski kapi in njihov učinek na incidenco padcev

“…Ponovna analiza padcev je bila opravljena v letu 2009 (Šavli & Trtnik, 2010), januarja 2010 pa je bila izvedena anketna raziskava med zdravstvenimi delavci (Vrabič, 2011). Po analizah je bilo septembra 2010 uvedeno ocenjevanje tveganja za padec ob sprejemu z lestvico tveganja (Nakagawa, et al, 2008) Nakagawa in sodelavci (2008). Analizo klasifikatorjev so avtorji izvedli s krivuljami ROC (Receiver Operating Characteristics Curve) ter z izračunom površine pod krivuljo (AUC) primerjali točnost modelov, pri čemer so izbrali najbolj ustreznega s senzitivnostjo 0,70 in specifičnostjo 0,69.…”