Design

This was a prospective cohort study conducted as a part of a broader, mixed methods investigation of psychological morbidity in older adults who have been discharged from hospital (Lalor et al., Reference Lalor, Brown, Robins, Lee, Connor and Russell2015). It included a baseline measurement of demographics, recent falls, depressive symptoms, and participation in physical activities taken just prior to their discharge from hospital (denoted as t_{n =0}). Repeated measures of depressive symptoms, participation in physical activities, self-reported physical capacity, and self-reported falls in the last month were taken at one monthly intervals from the baseline survey with follow-up extending to six months (time points denoted as t_{n =1,2,3,4,5,6}). A range of other outcomes relevant to the broader study was also taken at baseline, for example, measures of social isolation, quality of sleep, management of falls, and depression. This paper focuses on the temporal relationships between falls, participation in physical activities, and depressive symptoms post-discharge.

Participants

Participants in this study were older adults aged ≥65 years, who were admitted to subacute or inpatient rehabilitation wards of Monash Health and Peninsula Health in Victoria, Australia between January 2013 and September 2014. Older adults who had at least a two-week length of stay in hospital were eligible to participate. A minimum of two weeks were selected as it was reasoned that the surgery or illness that precipitated this hospitalization would have been a significant health event likely to have ongoing ramifications for the patient's health and function during the six-month follow-up period post-discharge. We excluded patients with significant cognitive impairment (defined as a score of >13 on the six-item Cognitive Impairment Test) (Brooke and Bullock, Reference Brooke and Bullock1999), hearing or speech impairment sufficient to prevent a conversation, or a discharge destination following hospitalization that was another hospital or a nursing home. There were no exclusion criteria based on English speaking ability. Accredited interpreters were employed as necessary to assist with consent and survey conduct.

Survey instruments

The “Behavioral Model,” as outlined by Fiske et al. (Reference Fiske, Wetherell and Gatz2009), was used to guide selection of survey items and questionnaires. The follow-up assessments measured self-reported falls in the last month, self-reported physical capacity, depressive symptoms, and participation in recreational and household physical activities. The three- and six-month assessments contained additional questions to collect data on outcomes that were related to the broader study.

The Geriatric Depression Scale-Short Form (GDS-SF) (Sheikh and Yesavage, Reference Sheikh, Yesavage and Brink1986) was used to assess symptoms of depression. The sensitivity and specificity of this test ranges from 79 to 100% and 67 to 80%, respectively, in identifying clinical depression in older adults (Lesher and Berryhill, Reference Lesher and Berryhill1994). A higher score represents greater severity of depressive symptoms. A score >5 is suggestive of depression and those with a score >10 almost always have clinically diagnosed depression.

The Phone-FITT household and recreational subscales (Gill et al., Reference Gill, Jones, Zou and Speechley2008) were used to measure participation in household and recreational activities in older adults. This brief survey has been shown to have a high test–retest reliability and moderate validity compared to accelerometry-based measures. The Phone-FITT household or recreational scores are the summative values of self-reported frequency and duration for each activity in the household or recreational activity category. A higher score means a higher level of physical activity.

Closed questions were used to collect data on the occurrence of self-reported falls in the last month. A fall was defined as an event which resulted in the person coming to rest inadvertently on the ground, floor or other lower level (WHO 2017). Participants were asked to self-report their physical capacity using a Likert scale of “can do without difficulty,” “can do but with difficulty,” and “cannot do without someone else” on four physical activities: getting on and off public transport without assistance from someone else; walking up and down stairs without a handrail or assistance from someone else; bending and picking up an object from the floor without the assistance from someone else and moving about their usual place of residence without assistance from someone else. A range of other demographic details were collected from participants at baseline, including pre-existing history of depression.

Procedure

The study was approved by Monash Health and Peninsula Health Human Research Ethics Committee. Consent to participate was obtained from hospital patients who met the inclusion criteria.

Interviewers were trained by the chief investigator T.P.H. in the administration of the surveys. The interviewers were allied health professionals who had not been involved with treatment of the participants. Questions and relevant prompting materials of the surveys were loaded onto an online survey program (www.surveymonkey.com). The participants were interviewed face-to-face in the hospital within 48 h prior to their planned discharge at baseline t_{n =0}, and followed-up at their residence at three months t_{n =3} and six months t_{n =6} post-discharge. A telephone survey was conducted at one, two, four, and five months t_{n =1},₂,₄,₅ post-discharge. Survey data were entered directly by the interviewers into Survey Monkey at the time of the interview using tablet computers. Regular meetings were held between the chief investigator and the interviewers to address problems that may occur during the administration of surveys to ensure uniformity in data collection.

Statistical analyses

Differences in the baseline data were compared between all participants who completed the survey at t_{n =0} and those who completed the survey at t_{n =6} to determine if a particular characteristic measured at baseline predicted completion of the six month follow-up assessment. Logistic regression was conducted using the dichotomous variable of whether the participant completed the survey at t_{n =6} as the dependent variable and baseline characteristics as independent variable (Table 1).

Table 1.

Baseline characteristics of study cohort at t_{n =0} and their differences with participants who completed survey at t_{n =6}

* Statistical significance at p < 0.05 from logistic regression with odds ratio OR and 95%CI.

^a 0–28, higher scores are more impaired.

^b Criteria (any2/3): (i) speaks a main language other than English at home, (ii) born overseas, (iii) identify themselves with a specific culturally group other than Australian or as an indigenous Australian or Maori.

^c Collected as part of EuroQol EQ-5D-5L questionnaire.

^d Depression diagnosed by general practitioner at baseline.

^e Geriatric Depression Scale-Short Form score 0–15, scores>5 is suggestive of clinical depression.

^f PhoneFitt household subscale measures household activities participants did over the last week.

^g PhoneFitt recreational subscale measures recreational activities participants did over the last week and seasonal activities over the last month.

The low household physical activity score at baseline reflected low activity because the participants were in the hospital when this data were collected.

The higher recreational physical activity score at baseline reflected higher activity because the participants were unlikely to be in hospital over the last month for seasonal activities when this data were collected.

The structure of the analyses undertaken was aimed at investigating three overall questions asked in a pairwise manner of the constructs of depression, falls, and physical activity. To illustrate these three questions would and their corresponding analyses, we selected the pairing of depression and falls. We first wanted to know if falls and depression were associated with each other. We, therefore, examined whether being a faller over the previous month (measured at t _n) was associated with depression symptom scores measured at the same time point (t _n). Depression symptom scores were used rather than a cut-off score for depression to capture the variation in depressive symptoms at each month. Next, we wanted to know whether having falls preceded a worsening of depression symptoms, thus indicating a temporal (potentially causative) relationship. We, therefore, examined whether being a faller over the previous month (measured at t _n) preceded a worsening of depression symptoms over the subsequent month (t_n −t_{n +1}). Finally, we wanted to know whether having symptoms of depression preceded subsequent falls, thus indicating a temporal (potentially causative) relationship. To address this, we examined whether depression scores (measured at t _n) preceded falls over the subsequent month (measured at t_{n +1}). Logistic regression analyses with data clustered by individual participant utilizing robust standard errors (Freedman, Reference Freedman2006) were employed to examine the first and third questions (with fall as a dependent categorical variable), while linear regression analysis with data clustered by individual participant utilizing robust standard errors was employed to examine the second question (with change in depression score as a dependent continuous variable). These analyses were conducted using outcome data collected across t_{n =1-6}, excluding baseline data to examine the patterns of association post-discharge from hospital.

Analyses examining the relationships between the paired constructs of falls and physical activity were conducted using the same approach. Physical activity was divided into two domains, household and recreational, and analyses were conducted separately for each. This is consistent with the published factor structure of the Phone FITT questionnaire (Gill et al., Reference Gill, Jones, Zou and Speechley2008). The same pattern of analysis was employed to explore the relationships between the paired constructs of depression and physical activity. The association between depression and physical activity was examined using scores for each construct collected at the same time point (t_n ). Examinations of potential temporal relationships were undertaken using depression scores collected at one time point (t_n ) and change scores between that time point and the subsequent one (t_n −t_{n +1}) for physical activity, and vice versa. Again, physical activity scores were divided into separate analyses for household and recreational domain scores, and linear regression analyses with data clustered by individual participant utilizing robust standard errors were employed.

The alpha criterion for statistical testing was set at p < 0.05. Data analyses were undertaken using STATA version 13 (Stata Corp, 2013). Analyses were repeated adjusting for the baseline covariate of pre-existing diagnosis of depression and self-reported physical capacity from t_{n =1-6} to examine whether these confounded the associations observed.

Transformation for depression scores (GDS-SF), physical activity household, and recreational scores were examined to normalize the data. The regression analyses were re-run with the transformed version of the data. The results obtained were then compared with the original analyses to determine if they have changed the results in any way.