Study Population, Design, and Setting
This prospective cohort study used data from the Japanese Dialysis Outcomes and Practice Patterns Study (J-DOPPS). It is a part of the Dialysis Outcomes and Practice Patterns Study (DOPPS), which is an international longitudinal study conducted on hemodialysis patients. The patients included in the J-DOPPS were randomly selected from some representative dialysis facilities in Japan. Their demographic information, laboratory data, comorbidities, dialysis conditions, medication (assessed every four months), and information on hospitalization and death were collected. All patients provided written, informed consent at the time of study enrollment. More details on the DOPPS are available in the literature.12
In this study, we included 2,442 patients aged 40 years, who participated in J-DOPPS phase V (2012–2015). As we tried to assess the decline in BADLs as our primary outcome, we excluded all patients who had already lost BADLs as was operationalized by dependency in at least three of the five BADLs at baseline. This criterion was used, as such patients would not be likely to be considered at-risk for a further decline in BADLs. In addition, patients for whom changes in ADLs were evaluated for less than 90 days, in which the change in ADLs will not occur sufficiently, were also excluded.
Measurements
Hospitalization
We identified the occurrence of hospitalization based on the participants’ medical records. Only those hospitalizations with more than a two-day hospital stay were identified because one-day hospitalizations were assumed not to influence the outcomes. We recorded the length of hospital stay for each hospitalization and the number of hospitalizations for each patient. The cumulative length of hospital stay and number of hospitalizations were used as exposure variables in the statistical analysis.
Change in Activities of Daily Livings
We evaluated BADLs and IADLs using self-report questionnaires. BADLs were assessed using the Katz index.13 Using this tool, participants answered whether they could independently perform five tasks (i.e., they answered either independent or not for each task). IADLs were assessed using the Lawton-Brody IADL scale,14 which asks participants to evaluate their ability to perform eight tasks on a 3-point scale (need no help, need some help, or unable to do at all). BADLs and IADLs were assessed twice: while registering for the J-DOPPS Phase V (baseline) and during follow-up the next year. The outcomes were defined as a decline in any one of the five BADLs and eight IADLs from baseline to follow-up.
Other variables
We collected information on the participants’ age, sex, body mass index (BMI), smoking behavior, and dialysis vintage (years on hemodialysis, from initiation to the baseline survey). BMI was categorized into <18.5 kg/m2, ≥18.5 to <25 kg/m2, and ≥25 kg/m2. We also obtained information on the presence of comorbidities (e.g., diabetes, cerebrovascular diseases, coronary heart disease, other cardiovascular diseases, congestive heart disease, cancer other than skin cancer, neurological disease, peripheral vascular disease, dementia, and psychiatric disorder) from participants’ medical records. Further, the most recent laboratory data on albumin, phosphorus, creatinine, single-pool Kt/V, pre-dialysis blood urea nitrogen, and hemoglobin levels were obtained at baseline. Subsequently, we calculated a functional status score at baseline by combining the scores of the Katz index and Lawton-Brody IADL scale using an algorithm developed in a previous study.2,15
Statistical Analysis
We conducted the following statistical analyses only for patients without any missing data (i.e., complete case analysis). All statistical analyses were conducted using Stata 15.1 (StataCorp, College Station, TX).
In the descriptive analysis, we described the participants’ baseline characteristics according to their hospitalization status; we used means and standard deviations (SDs) or medians and interquartile ranges (IQRs) for continuous variables and the number and proportion for categorical variables. The distribution of cumulative length of hospital stay among the patients who had been hospitalized was depicted in a histogram. The proportion of decline in each BADL and IADL according to hospitalization status was depicted in a bar graph.
To clarify the association between hospitalization and declines in ADLs, we calculated the risk ratio (RR) and 95% confidence interval (CI) from the mean predicted probabilities based on a fitted logit model, in which the estimated coefficients are transformed into probabilities through a logistic function. The RR can then be calculated as the ratio of the estimated probabilities, using the user-written command “adjrr” in Stata.16
In the primary analysis, the cumulative length of hospital stay was set as the independent variable. Then, we separately calculated the RRs for one-day increments of hospital stays and 95% CIs for the decline in BADLs and IADLs with adjustments for potential confounding factors (i.e., age, sex, dialysis vintage, BMI, functional status score, comorbidities, albumin, phosphorus, creatinine and single-pool Kt/V). We used cluster-robust variance to consider cluster effects according to the facility. For a simple interpretation of the impact of the exposure on the outcomes, we compared the RRs of 7- and 30-day increments for hospital stays with 10-year increments for age and having diabetes. A restricted cubic spline curve analysis with three knots was used to visually confirm the linear or nonlinear relationship between the cumulative length of hospital stay and predicted probabilities of decline in BADLs and IADLs. For this analysis, we employed logistic regression models, with adjustment for the same confounding factors as in primary analysis.
For the secondary analysis, we first categorized the number of hospitalizations into three categories (0, 1, or ≥ 2) and used it as an independent variable (with 0 as the reference) in the same statistical model as was used in the primary analysis. Second, we conducted the aforementioned two analyses by age group (< 65, ≥ 65 years) to verify whether the magnitude of the association between hospitalization and decline in ADLs differs between older and younger patients. To test the interaction, we added the product term for the cumulative length of hospitalization and age group for the primary analytic model and that of the number of hospitalizations and age group for the secondary analytic model. The statistical significance of the product terms was evaluated using a Wald test. Third, we conducted analyses by the two dialysis vintage groups that were classified based on ≥ 5 years (the closest value to the median) or < 5 years, as well as the age group. Fourth, we conducted an analysis using infection-, cardiovascular disease- and vascular access-related hospitalization as the exposure variables. We evaluated the associations between these cause-specific hospitalizations and a decline in ADLs.
Finally, we conducted three sensitivity analyses to confirm the robustness of the primary analysis. First, we defined the outcomes as a decline in two of the five BADLs and eight IADLs. Second, since the different evaluation periods for the change in ADLs for each patient (ranging from 93 to 566 days) could affect the results, we analyzed only those patients with an evaluation period within the IQR (338 to 376 days). Finally, we included 148 patients who were previously excluded due to dependency as measured by having at least three of the five BADLs during the primary analysis. For all analyses, p < 0.05 was considered statistically significant.