Between Jun-2018 and Aug-2020, a cross-sectional knowledge study was conducted in a subset of patients enrolled in the CirCare study (Fig. 1). Briefly, CirCare is a prospective multicentre longitudinal study of 581 patients with cirrhosis who were consecutively recruited when they attended liver clinics or were admitted to one of five hospitals in Brisbane and Logan, Queensland, Australia between Jul-2016 and Dec-2018. Patients with cirrhosis were eligible for the ‘knowledge survey’ if they had agreed to be recontacted for future studies, were not known to be deceased at the time the survey was posted, and if their last contact with researchers was after 1 October 2017 (to ensure relatively recent recruitment). A total of 292 surveys were mailed to eligible patients with instructions to return the completed anonymised survey using a prepaid envelope. A reminder was sent out by mail and, for those patients with telephone details available, a follow-up telephone call was attempted. Aboriginal and Torres Strait Islander patients (also referred to as Indigenous Australians) were provided with a second opportunity to participate in the ‘knowledge survey’ while taking part in a qualitative sub-study during April-July, 2020. Patient outcomes from the date of CirCare recruitment to Dec-2019 were obtained via data linkage. As data regarding date and cause of death were provided in Jun-2020, this information was not available when the knowledge survey was distributed.
The ‘knowledge survey’ (see Supplementary Table 1) included 17 questions derived from prior studies.[13, 14] An expert panel including two hepatologists, a liver nurse, and a liver pharmacist (EEP, KS, PW, and KH) subsequently reviewed the questions (blinded to the data) and unanimously deemed eight to be “key knowledge” about liver disease. Patients’ knowledge scores were calculated by assigning correct responses a score of 1, and incorrect or ‘I don’t know’ responses a score of 0. The proportion of correct answers were calculated for each patient and referred to here as the “key knowledge” score. The total score, over a range of 0 to 100%, was dichotomised using the median score as the cut-off point. Patients were categorised as ‘good knowledge’ for scores ≥ 62.5% (that is 5 out of 8 correct answers; Fig. 2), and ‘poor knowledge’ if < 62.5%.
Health Related Quality Of Life
Health-related quality of life data were self-reported via a face-to-face interview at recruitment into the CirCare study. The Short Form 36 (SF-36), a widely used and validated quality of life tool, includes 36 questions grouped into 8 domains (general health, physical functioning, social functioning, bodily pain, role limitations due to physical problems (role physical), emotional wellbeing, role limitations due to emotional problems (role emotional) and vitality). Health-related quality of life raw domain scores were transformed to range from 0 to 100, with a higher score indicating a higher health status.
Health service use data were obtained via data linkage from the Queensland Hospital Admitted Patient Data Collection database and the Emergency Data Collection database that contain information on all hospital episodes of care for patients admitted to Queensland public and private hospitals. Hospital admissions were categorised as ‘cirrhosis admissions’ based on recorded ICD-10‐AM codes as previously described. The accuracy of this algorithm for identification of patients with cirrhosis has been reported to have an 88% positive predictive and 76% negative predictive value. Emergency presentations were categorised as ‘cirrhosis-related presentations’ if they had a primary or other diagnosis of cirrhosis, cirrhosis-related diagnosis, or cirrhosis-related complications; namely chronic hepatic failure, portal hypertension, hepatorenal syndrome, spontaneous bacterial peritonitis, ascites, variceal bleeding, hepatic encephalopathy, jaundice, or alcohol related presentation (e.g. alcoholic hepatitis, alcoholic liver disease unspecified). Death Registration Data and Cause of Death Unit Record File were obtained from the Australian Bureau of Statistics. The National Hospital Cost Data Collection database was the primary source of cost data for all hospital admissions at public and private hospitals. Costs included aggregated direct plus overhead costs. Hospital admissions, emergency presentations and death data were available from the CirCare study recruitment date to Dec-2019. Cost data were available from the CirCare study recruitment date to Jun-2018. A diagram showing the timeline for recruitment and collection of outcome measures is displayed in Fig. 1.
Sociodemographic And Clinical Characteristics
Sociodemographic data were self-reported at recruitment into the CirCare study. Place of residence was categorised according to rurality of residence and the Index of Relative Socioeconomic Advantage and Disadvantage. Clinical information at the time of recruitment was extracted from patients’ medical records. Severity of disease was classified using the Child-Pugh class and by absence (compensated cirrhosis) versus presence of cirrhosis complications (e.g. ascites, hepatic encephalopathy). Comorbidity burden was measured using the Charlson Comorbidity Index using validated coding algorithms.
Data analyses were conducted using Stata/SE (version 15; Stata Corporation, College Station, TX). Descriptive analyses of patient characteristics and responses to the knowledge survey were presented as frequency (percentages) and mean (standard deviation, SD).
Multivariable logistic regression analysis reported odds ratios (ORs) with associated 95% confidence intervals (CIs) to examine factors that were independently associated with “key knowledge”. The decision as to which independent variables were included was first determined based on the results of bivariable analyses. We then ran multivariable analysis to appreciate the extent of confounding and applied stepwise model selection (p = 0.20 as the significance level at which variables were entered to or removed from the model). As longer duration of disease may provide patients with more opportunities to receive information about cirrhosis, we have adjusted the estimates for duration of disease. The final model for “key knowledge” included the following covariates: presence of complications of cirrhosis, age, education level, and duration of cirrhosis.
The rate of hospital admissions and emergency department presentations was calculated using person days at risk (PDAR) as a denominator. Cases were followed from date of recruitment date to CirCare until death or December 31, 2019, whichever came sooner. Poisson regression was used to compare rate of admission or emergency department presentation according to “key knowledge” status (incidence rate ratios (IRR) and 95%CIs were reported). Education level, socioeconomic status, duration of cirrhosis and severity of liver disease (measured using presence of cirrhosis complications) were included in the model. As Child-Pugh score was unavailable for 3 patients, the presence of cirrhosis complications was included in multivariable analysis as a marker of severity of liver disease.
We reported IRRs to describe the ratio of costs of hospital admissions according to “key knowledge” status. PDAR included data from the CirCare study recruitment date until the date of death or June 30, 2018, whichever came sooner. As comorbidity burden and severity of cirrhosis are associated with health service use, we have included Charlson Comorbidity Index and presence of complications of cirrhosis in the model, along with education level.[29, 31]
Sensitivity analyses were carried out by: (1) using a cut-off of 57.1% on the 8-item score (58.5% of study participants had ≥ 51.7% of correct answers; Fig. 2); and (2) by including three extra items deemed to be “key knowledge” of liver disease by half of the experts and using a cut-off of ≥ 60.0% (median score of the 11-item survey).
Cumulative overall survival estimates according to patient knowledge were calculated using the Kaplan–Meier method (log-rank statistic). All cases were followed from the CirCare study recruitment date until date of death or December 31, 2019, whichever came sooner. Multivariable Cox regression analysis reported in terms of hazard ratios (HRs) with associated 95%CIs was used to assess the differences in survival according to “key knowledge” status. The vce(robust) option was used to obtain robust standard errors for the parameter estimates to control for mild violations of underlying assumptions. All p-values were 2-sided.