3.1 Study source

Data was extracted from the Dryad Digital Repository (https://datadryad.org/), which allowed users to freely download the raw data. We cited the Dryad data package ((Hayakawa et al., 2014a, 2014b), “Data from: Risk factors of falls in inpatients and their practical use in identifying high-risk persons at admission: Fukushima Medical University Hospital cohort study,” Dryad, Dataset, 10.1136/bmjopen-2014-005385) in the present study.

3.2 Study design and participants

This cross-sectional study was conducted between August 2008 and September 2009 among 8,598 inpatients (aged ≥ 18 years) included in the Fukushima Medical University Hospital (FMUH) database. This was a study at FMUH (Fukushima, Japan), designed to (1) identify the risk factors for falls in a hospital setting and (2) use such factors to identify the risk of falls in high-risk persons.

The details of the FMUH study were described previously (Hayakawa et al., 2014a, 2014b). Briefly, 9,957 participants who participated in the program between August 2008 and September 2009 were recruited at FMUH and were followed up until discharge. We excluded participants who were lost to follow-up (N = 487), with age <18 years (N = 20), and with a hospitalization time >40 days (N = 628). Participants with missing covariate information were excluded from the analyses (N = 224). A total of 8,598 participants were included in the final analysis (Figure 1). Our study was exempt from institutional review board oversight, because the Dryad database is accessible to the public, and the patients in the database are de-identified.

3.3 Variables

The primary outcome measured was a fall event, which was treated as a classified variable (no or yes). The primary exposure of interest was the hospitalization time, which was a continuous variable.

The covariates included in the analysis were age, sex, history of falls, cognitive dysfunction, planned surgery, wheelchair use, assistance in movement, eye sight, use of a remote caring system, impaired manual muscle test (MMT, defined as MMT <4) (Bohannon, 2005), assistance with activities of daily living (ADL), medication use (for example, laxative, sedative, hypnotic, and psychotropic medications).

Information was collected using a structured questionnaire through face-to-face interviews and clinical records at admission, conducted by nurses and doctors. A low level of ADL was defined as the need for help with any one of the seven activities (standing, sitting, dressing, eating, toileting, evacuation and washing of face).

3.4 Statistical analysis

The continuous variable, hospitalization time, was defined as a categorical variable. All the normally distributed and skewed continuous variables were described as mean (standard deviation) or median (interquartile range), and categorical variables were described as frequencies (%). The baseline characteristics of the different hospitalization time groups were analysed using one-way analysis of variance (normal distribution), the Kruskal–Wallis H (skewed distribution), and chi-square test (categorical variables).

Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for falls associated with the hospitalization time using four logistics proportional hazards models as follows: (1) unadjusted; (2) adjusted for age and sex; (3) adjusted for history of falls, MMT (left foot), MMT (right foot), cognitive dysfunction, eyesight, assistance with ADLs based on Adjust I; and (4) adjusted for planned surgery, wheelchair use, assistance with movement, laxative use, use of a remote caring system, sedative use, hypnotic medication use and psychotropic medication use based on Adjust II.

We determined whether each potential confounding variable was associated with falls. A confounder was defined as a variable with a clinically meaningful relationship with a risk factor and outcome (regardless of statistical significance). Besides, we also used restricted cubic splines with four knots at the 5th, 35th, 65th and 95th centiles to flexibly model the association between hospitalization time and fall incidence. Subgroup analyses were conducted by age, sex, history of falls, cognitive dysfunction, and laxative, hypnotic and psychotropic medications use using stratified logistic regression models. All analyses were performed using the statistical software packages R 3.3.2 (http://www.R-project.org; The R Foundation) and Free Statistics software version 1.2. p-Values <.05 (two-sided) were considered statistically significant.

4.1 Baseline characteristics of study participants by categories of hospitalization time

Figure 1 shows the number of eligible participants and participants analysed in this study. Participants who were lost to follow-up, with age <18 years, with a hospitalization time >40 days, and those who lacked clear records on covariants were excluded. Therefore, 8,598 participants were included in the final analysis. Inpatients who met the study criteria were identified in the FMUH database from August 2008 to September 2009.

Table 1 Of the 9,957 participants, 1,359 were excluded, and thus the data from 8,598 inpatients (mean age 59.30 ± 17.39 years; 4,294 male, 4,304 female) were analysed in the current study. The Incidence rate of falls was 1.81% (N = 156) in the present study, then the participants were divided into tertiles based on the hospitalization time. The participants with the longest hospitalization time were likely to fall, compared with those who had the shortest hospitalization time (p < .001). The mean age, sex, number of falls, history of falls, wheelchair use, assistance with movement, MMT right or left, laxative medication use, hypnotic and psychotropic medication use, cognitive dysfunction, eyesight and assistance with ADLs, among others, varied significantly between groups. Meanwhile, the proportion of participants who fell was higher in the participants using hypnotic and psychotropic medications.

4.2 Association of the hospitalization time with fall incidence

Table 2 shows the association between the hospitalization time and falls using the multiple logistic regression models. In the non-adjusted model, there was an initial decrease followed by an increasing risk for developing falls as the tertile of hospitalization time increased; participants who had a hospitalization time in the highest tertile versus the lowest tertile had an approximately twofold increased risk of falls (OR, 1.95; 95% CI, 1.35, 2.81), p for trend <.0001. Similar results were obtained in the model adjusted for all covariates, including age, sex, history of falls, MMT left, MMT right, cognitive dysfunction, eyesight, assistance with ADL, planned surgery, wheelchair use, assistance with movement, laxative medication use, remote caring system, and sedative, hypnotic and psychotropic medications use (OR 1.57, 95% CI 1.07 to 2.30, p = .021), p for trend = .006.

4.3 Inflection point analysis of the hospitalization time on fall incidence

The relationship between falls and the hospitalization time was non-linear (p value, non-linear <.001), and a model using restricted cubic splines with three knots revealed a U-shaped association, with a decreasing fall incidence as the hospitalization time increased from 0 to 8 days (OR 0.72 [0.62 ~ 0.83], p < .001); beyond 8 days, the fall incidence increased as the hospitalization time increased (OR 1.06 [1.04 ~ 1.09]) (Figure 2, Table 3). In the figure, the blue solid line indicates the estimated risk of fall incidence, and the light blue shadow represents point-wise 95% CI adjusted for age, sex, history of falls, MMT (left foot), MMT (right foot), cognitive dysfunction, eyesight, assistance with ADLs, planned surgery, wheelchair use, assistance with movement, laxative medication use, remote care system use, and sedative, hypnotic and psychotropic medications use.

4.4 Subgroup analysis

Sex, history of falls, cognitive dysfunction, and laxative, hypnotic and psychotropic medications use are known confounders of hospitalization time–fall association (Hayakawa et al., 2014a, 2014b). In order to investigate whether different layers affected the hospitalization time and falls, stratification analysis in subgroups was performed (Figure 3). We found a stable correlation between the hospitalization time and falls.

5.1 Strengths and limitations

The strengths of our study include the fact that our sample size was relatively large compared to that of previous similar studies performed on inpatients (≥18 years). Nevertheless, there are some limitations to our study. First, our research participants were restricted to inpatients (≥18 years); therefore, there was a certain deficiency in the universality and extrapolation of the research. Second, there was a selection bias in the present study. The hospitalization time of our participants was ≤40 days, which might have led to an underestimation of the incidence of falls. Third, the data lacks disease type, patients' clinical conditions and worsening or improvement during the hospital stay, treatments received, and characteristics of the care team, but it is important to evaluate preventive factors against falls. In the future, we need a large sample of forward-looking fall data to improve the research in this area.