Health-related quality of life in family members of patients with an advanced cancer diagnosis: A one-year prospective study

Design

This was a prospective quantitative study designed to follow mental and physical dimensions of HRQOL in family members of persons with advanced lung or gastrointestinal cancer over a 1-year period, beginning 3 months after diagnosis. The study was conducted during 2007–2011. Data were collected on five different occasions, 3, 6, 9, 12 and 15 months after the patient’s diagnosis, and are termed assessments A1, A2, A3, A4 and A5, respectively.

Research ethics

Ethical approval for this study was granted by the research ethics committee at Linköping University, Sweden. Informed written consent was obtained from all participants prior to the study. It was made clear to them that participation was voluntary, and they were free to withdraw whenever they wished without any consequences related to care for themselves or the cancer patient. Confidentiality was guaranteed, and the findings could not be linked to the individuals.

Selection of the study group

Family members incl`ded in this study had a sick relative who up to 3 months earlier had been diagnosed with advanced lung cancer or cancer in the upper gastrointestinal tract at either one medical clinic or one of the two surgical clinics at two hospitals in the south of Sweden. Eligible family members had to be aged 18 years of age or older and able to speak Swedish.

Initially, 200 patients with cancer were asked by one of eleven nurses or one of two physicians at the three clinics if they were willing to give written information about the study to their closest family member and ask them to participate (Figure1). If patients accepted, they received two letters indicating the purpose and design of the study, one to themselves and one to the chosen family member. Sixty-four family members agreed to participate, but seven of them withdrew before the study began because the patient was too ill. Demographic data on those who withdrew prior to the start of the study (n = 7) were not available. Twenty-one family members dropped out because the patient died or became too ill. Two types of calculation were used to analyse differences between the study group and drop-outs. First, any significant differences in demographic data (age, gender, education, occupation or relationship) were examined (Table1). Second, differences in HRQOL between the groups were explored at baseline. No significant differences were found between the study group (n = 36) and drop-outs (n = 21) in either calculation.

Data collection procedure

Family members who gave their written consent to participate were contacted by phone by the first author. They were then mailed two HRQOL self-report instruments together with a cover letter every 3rd month. To reduce the number of drop-outs, they were also phoned 1 week before each data collection to remind them that the questionnaires were going to be mailed again. A similar reminder was provided 2 weeks after the mailed questionnaires if their response had not reached the first author.

Instruments

The two self-report HRQOL instruments used in this study were the Short Form Health Survey (SF-36) and the EuroQol (EQ-5D). These instruments were selected because they are considered complementary in respect to different variables of similar dimensions.

Health-related quality of life (HRQOL)

The SF-36 is one of the most widely used self-assessment instruments for measuring HRQOL. It consists of 36 items making up eight scales that measure a physical and a mental health dimension. The physical health dimension, represented by the Physical Component Summary Score (PCS), contains the scales Physical Functioning (PF), Role Physical (role limitations due to physical problems; RP), Bodily Pain (BP) and General Health (GH). The mental health dimension, represented by the Mental Component Summary Score (MCS), contains the scales Vitality (VT), Social Functioning (SF), Role Emotional (limitations due to emotional problems; RE) and Mental Health (MH)[23]. Raw scores were coded, re-calibrated, summated and transformed from 0 to 100 for each item, with higher scores reflecting better HRQOL, following norm values for the general Swedish population. The eight scales and the two dimensions, PCS and MCS, were calculated according to the standard SF-36 algorithms in the test manual[23, 24].

The EQ-5D, also a self-assessment instrument, comprises five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. The answers were scored by an algorithm from the EQ-5D manual[25, 26], an index value of HRQOL with higher scores reflecting better health status. The respondents were also asked to draw a line on a visual analogue scale (VAS) thermometer that ranged from the worst imaginable health state today (0 points) to the best imaginable (100 points). The index value of the EQ-5D can be generated from the dimensions with a range of −0.594 to 1 by applying scores from standard population values[27].

Analysis

Family members’ characteristics in the drop-out group compared with the study group were examined using the chi-square test. Linear mixed models with repeated observations and method restricted maximum likelihood were applied to analyse changes between 3 months and 15 months for family members in separate comparison analysis of each value, with measure 1 at baseline (3 months after the patient’s diagnosis). The variable of age was dichotomised on equal percentiles in two age groups (20–65 and 66–84 years). Because the study group was small, two equally sized groups were preferable for statistical comparisons. Moreover, such classification meant that the older group came to consist mainly of retired individuals, and the younger group comprised mainly people who were economically active. SF-36 dimensions were analysed with the Mann–Whitney U Test for independent samples between age group and between family members’ relationship to the patient. Although the analysis was non-parametric, the family members’ HRQOL on the SF-36 and EQ-5D are presented in the tables as means, medians and standard errors of the means to facilitate comparisons to previous studies. The associations between age and relationship to the patient with cancer (partner or grown child) were calculated using Spearman’s correlation coefficient. A stepwise logistic regression analysis using the forward Wald method was conducted on all five assessments with the eight scales in the SF-36 as independent variables and age and relationship as dependent variables. The aim was to describe which scales had the greatest effect by age and relationship. The significance level was assumed at α = 0.05, and the statistical calculations were carried out in SPSS, version 19.1. Statistical differences were tested between norm-based scores for the SF-36 and EQ-5D and respective mean scores at A1–A5 using Statistica, version 10. Because age may be an important factor in HRQOL, it was important to get a comparable age structure in a comparison of scores of the study group and norm-based scores.

Methodological considerations

The study population included relatively few family members and resulted in low power in detecting small changes over time and differences between subgroups in the population. This was the case, despite our recruitment of family members from two medium-sized hospitals and a rather long inclusion period (2007–2010). There were drop-outs both before the start of the study and during the study. The selection of family members was accomplished through assistance by staff (nurses and physicians) who initially asked the patients and secondly by the patients who asked their family member about participation. This procedure explains why there is no information about the reasons for drop-outs and demographic data of the initial 200 family members or the seven that dropped out right after consenting to participate. Despite several reminders provided during the data collection procedure to prevent drop-outs, the initial recruitment could have been done differently. The initial number who dropped out might have been reduced if the family members had been approached personally by the researcher or only one responsible nurse at each hospital.

Of the 57 family members entering the investigation, a total of 36 completed the study with participation in all five data collections. Analysis of drop-outs showed that the 21 family members who dropped out during the study were not significantly different in age, gender, education, occupation or relationship from those who completed the study. In addition, a strength of the data is that there was no difference in HRQOL or EQ5D at baseline between drop-outs after the start of the study and the study group who participated in all five data collections, which reduces the chance that the study was not representative. However, the large non-response rate makes uncertain the study’s generalisability, which must be kept in mind when interpreting the results.

Another reason for the low statistical power[47] was the authors’ decision to include only family members who filled in the questionnaires in all five data collections. The authors assessed it as better if the same individuals were included in the comparison over time than comparing groups of individuals in which the size decreased because of drop-outs. Seeing changes over time can result in more reliable data. Even though the power could have increased if all individuals were included, uncertainty would increase because there was no single group. The drop-outs during the study were explained by increased sickness in the patients, a factor that was out of our control during the year-long data collection. This reason for drop-outs has been previously described in the literature related to severe or advanced cancer patients[11, 21]. Therefore, the performed statistical analyses were less powerful, and logistic regression analysis should be considered from a descriptive perspective. In addition, the findings of the logistic regression analysis of type of relation on HRQOL were not included in the results because of a large difference in group size (partners, n = 28 and children, n = 8). The ability to predict values for the group of children was relatively low with percentage correct 50–62% compared with the analysis of partners (percentage correct 83%) and age (percentage correct 85%).

When a large number of comparisons are statistically examined, there is a risk of mass significance. With an alpha-level of .05, the risk for random significance is 1 in 20. One way to diminish the risk of mass significance is to lower the alpha level. Procedures used to calculate a new alpha level are Bonferroni’s test and the Dunn-Sidak correction[47]. However, in this study the authors determined that these tests would be too conservative and might increase the possibility of a type II statistical error. For this reason, weak or solitary significances should be considered with some scepticism. Furthermore, the collected data are ordinal-level and not equidistant, and thus there is a risk of bias in evaluating the size of changes[48] owing to the uneven distribution of such changes. Conclusions related to our results are therefore based on the patterns of significant differences that were in the same direction without deviant values. Linear mixed models can be used to describe nonlinear relationships across time in a longitudinal dataset with multiple missing data points. The strengths of the mixed models are the ability to accommodate missing data points and the ability to model nonlinear, individual characteristics. The mixed model emphasises patterns of change and individual differences and assumes not a normal distribution but rather systematic change.

The proportion of men in the study group of family members was low, which might be explained by the selection method. Both sexes were represented in nurses (females) and physicians (males) who asked about participation. However, 26 of the 36 patients who asked one of their family members to participate were males (72%), and 20 of them (77%) chose a female partner. Six of the males chose a grown child, and five of those were daughters. Cancer incidence and mortality in Europe predominately involves men with lung or upper gastrointestinal tract cancer[49]. These cancer diagnoses are the most common worldwide among men and constitute 42% of new cases and 48% of total cancer deaths. Given that a diagnosis of lung or upper gastrointestinal tract cancer is more common among men, it seems reasonable that our study group was predominately male cancer patients and female family members.

The longitudinal design[50] was appropriate for studying the dynamics of the variables over time given that cancer patients’ serious illness progresses rapidly and may affect the family members[50]. The 1-year follow-up period was considered appropriate in respect to investigating outcomes for family members dealing with the mortality of an advanced cancer patient.