Study design and data collection
We conducted a cross-sectional survey in Heilongjiang, a province in China with a medium-sized population (38.12 million in 2015) and economic products ($6386 per capita GDP in 2015) [34].
Three cancer centres were selected purposively because they were located in the capital city (Harbin) of Heilongjiang province serving as major referral centres for cancer patients. All of the three centres were affiliated to a tertiary hospital, providing specialist care to leukemia patients across the entire province. The investigators obtained permission from the participating hospitals to conduct the study and asked for a list of admitted leukemia patients over the period of data collection (July 2015 to February 2016). The eligibility of the participating patients for this study was assessed by their doctors and nurses. The participating patients had to have a dedicated primary family caregiver, this being the FC who provided their most of time to care without receiving any financial compensation. Then, 12 trained postgraduate research students were deployed to these centres to conduct face-to-face interviews using a structured questionnaire (Additional file 1). These interviewers did not have a service relationship with the participants. They approached the selected FCs, explained the purpose and protocol of the study, and sought written informed consent from the participants. The participants were encouraged to self-complete the questionnaire unless they requested assistance from the interviewers.
In total, 349 primary FCs were invited and 314 (90%) completed the questionnaire. Five returned questionnaires were excluded from the final analyses due to missing items that are essential for calculating health utility. This resulted in a final sample size of 306 (88% of the invited participants).
Measurements
Dependent variable - health utility
Health utility is a numeric index, with 0 indicating death and 1 representing perfect health. Usually, it is obtained using a generic HRQOL instrument [26]. In this study, we chose the EQ-5D-3L simply because it is the most commonly used instrument [11] and a Chinese population preference value set was recently made available [21].
The EQ-5D-3L contains five items measuring mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Respondents were asked to rate their current status and experience at three levels: no problems; moderate problems; extreme problems. Each of the combinations (a total of 243) of the five dimensions was given an index score based on a preference weight derived from the general population [21]. In the Chinese value set, the minimal preference weight is − 0.149, indicating a worse than death status, and the maximal preference weight is 1, indicating full health.
Independent variables
Health utility can be determined by many factors. In this study, we adjusted the health utility scores by the socio-economic characteristics of the FCs, such as age, gender, educational attainment, marital status, employment, household income, and relationship to patient.
Previous studies [22, 46] demonstrated that the characteristics of patients impose a significant impact on the need for family care and the level of emotional distress of the FCs. Our questionnaire captured the following data in relation to patient characteristics: age, gender, ethnicity, classification of medical insurance, time of diagnosis, and classification of leukemia. These characteristics were associated with how patients respond to their illness and the potential clinical outcomes of cancer treatments [5].
Workloads have been widely accepted as an important factor influencing HRQOL. High workloads can lead to stress, anxiety and depression [7]. In this study, we measured the average daily hours committed by the FCs for caring for the patient while in hospital and the overall annual load (months) of care. We used the Hospital Anxiety and Depression Scale (HADS) to measure the level of anxiety (7 items) and depression (7 items) of the FCs in the prior week. The level of anxiety or depression of FCs caring for leukemia patients was classified as severe (15–21 summed score), moderate (11–14 summed score), mild (8–10 summed score), or normal (0–7 summed score) [47].
Support from the family and community may alleviate the stress levels experienced by the FCs and subsequently improve their HRQOL [5, 18]. We measured the level of social support of FCs with the validated Social Support Rating Scale (SSRS), which resulted in a total score ranging from 66 to 0 [30, 42]. Respondents were divided equally into two groups: ‘high support’ or ‘low support’. We used the family APGAR (adaptation, partnership, growth, affection, and resolve) scale to assess the level of family support of FCs, which resulted in a total score ranging from 10 to 0 [8, 9]. Respondents were categorised into three groups for the purpose of statistical analyses. The summed score was graded as 0–3 (severely dysfunctional), 4–6 (moderately dysfunctional), and 7–10 (highly functional).
Data analyses
We reported the means and standard deviations (SDs) of the health utility scores of the FCs, as well as the medians and inter quartile ranges (IQs) of these scores. The distribution of the health utility scores measured by the EQ-5D-3L was biased, with 31.0% of respondents reporting the highest possible score of 1.
We compared the utility scores of the FCs with those of the local (Heilongjiang) general population using the Wilcoxon signed-rank test. Such a comparison was made for the following reasons: (1) Population norms were available from a representative sample of the local population in Heilongjiang as part of the fourth National Health Services Survey (NHSS) 2008, involving 15,875 individuals (from 5530 households) in 13 cities and counties [13]. (2) FCs came from this local population. (3) No comparable FCs for other patients were available.
The independent variables that were associated with the health utility of the FCs were identified through the Kruskal-Wallis analysis of variance (p < 0.05) and then entered into a multivariate median regression model (all independent variables were coded or transformed into categorical measurements). Ceiling effects are common in HRQOL studies [13, 37], including the EQ-5D-3L [2]. The literature recommends Tobit regression, censored least absolute deviations, and median regression to deal with data of such a censored nature [13, 14, 37], because they have theoretical advantages over the ordinary least squares estimator [13, 25, 38]. When censoring occurs in less than 50% of cases, median regression (robust to censoring, outliers and heteroskedasticity) is equivalent to censored least absolute deviations [25].
The findings of the median regression model were further confirmed by testing the difference in the prevalence of problems (moderate or extreme problems in mobility, self-care, usual activities, pain/discomfort, and anxiety/depression) in the FCs across different categories of the independent variables using chi-square or Fisher’s exact tests.
Data analyses were conducted using SPSS version 22 and STATA version 11, with a p value less than 0.05 being deemed as statistically significant.