Study design
The study was divided into two parts: 1) health states were developed in the first part of the study involving qualitative, semi-structured interviews with parents of children with primary HLH and clinicians; 2) a time trade-off methodology was used to elicit utility values for each of the health states from a broadly representative sample of the general public in the UK.
Ethics review
The study protocol was reviewed and approved by an independent review board: Salus IRB (date of approval: 14th August 2018).
Part 1: health state development
Health state descriptions or ‘vignettes’ were developed through a process that combined information gathered from consultation with clinical specialists and in-depth qualitative interviews with parents of children with primary HLH aged between 1 and 18 years old. This is a standardized approach which has been used in similar studies previously [7, 8]. Qualitative feedback obtained from interviews was used to describe patient’s experience of different states of disease and treatment as per each health state.
In depth interviews
Parents were recruited through patient support groups and social media such as Facebook pages in the United States (US), which were selected based on previous research in this area and study team. An advertisement was placed on the social media pages and support group newslettet to contact the team by email if potential parents were interested. Participants who contacted the team were then asked to schedule a convenient time to screen. If participants were eligible, they were then asked to schedule a convenient time for a telephone interview. Three parents were recruited from Facebook support page created by parents and one parent was recruited from a support group. Potential participants included people who reported that they were a parent or primary caregiver of a child (< 18 years) who had been diagnosed with HLH that required treatment with a stem cell or bone marrow transplant within the previous 10 years. All parents (N = 4) were from the US and all interviews were conducted in English. The parents talked about the hospital where their child received care and provided details that were consistent with their child having HLH. One round of interviews was conducted with parents and two rounds of interviews were conducted with clinicians. Clinicians (N = 3) were recruited who had experience of managing and treating patients with primary HLH. The clinicians were all known to the study sponsor. All participants provided written consent and then took part in a semi-structured, telephone interview. The parent interviews explored the circumstances which led to paitent’s diagnosis, patient’s experience of HLH symptoms, the treatment and burden that the disease has on his/her quality of life. The interviews lasted approximately an hour. Parents were also asked to describe the treatment and complications that their child had experienced.
The first round of interviews explored the impact of primary HLH and the types of treatment available and associated adverse events. Physicians discussed management of HLH and known adverse events. Based on the first round of interviews, draft health state descriptions were developed. These were validated in the second round of interviews with clinicians for accuracy. All Interviews were recorded and transcribed. Parents received a $50 Amazon voucher for their participation.
The information was summarized by one team member and reviewed additionally by a second team member, without any use of software. Key information from interviews with parents and clinical experts was compiled and summarised in terms of key areas of functioning (usual activities, emotional wellbeing, social functioning, cognitive ability), symptoms and quality of life. If symptoms, areas of functioning and aspects of quality of life were mentioned by more than one participant, they were included in the draft health states. For each state the intention was to capture sufficient information to be able to describe the symptom burden (pain, fatigue etc), the psychosocial impact and the impact on physical functioning. Interviews were summarised in terms of these different aspects of HRQL. Information was collected and included if it was reported by more than two parents and by at least one clinician. From these interviews, the draft vignettes were derived which represented a typical life cycle of treatment. These included active HLH, active HLH receiving chemotherapy, active HLH with neurological (CNS) involvement in the form of seizures, undergoing stem cell transplant, successful treatment or cure, graft versus host disease, and receiving end of life care.
Expert review
Two further expert interviews were conducted to review the content of the vignettes. Clinicians were asked to comment on how appropriate anad accurate each description was. Changes to the wording of the vignettes were suggested to improve the accuracy of the health states. These changes were incorporated and the final vignettes used for the valuation exercise are presented in the Appendix.
Part 2: health state vignette valuation
Members of the UK general public were recruited to take part using newspaper advertisements and an existing database of volunteers. The general public were recruited to provide a societal perspective on the impact of HLH. Societal perspectives are generally preferred by HTA bodies because they are making recommendations regarding the public’s access to health care. The sample was designed to reflect the general population in terms of age and gender. All interviews were conducted by trained interviewers. All participants gave written informed consent. Participants completed a brief socio-demographic questionnaire and the Time trade-off (TTO) interview. The TTO interview assesses the value or worth of different states of health by exploring how many years of life people may be willing to sacrifice in order to avoid such a state. It is generally preferred over other approaches (e.g. standard gamble) [6]. The TTO method is a preferred approach by assessment bodies such as NICE [6].
During the TTO exercise, all participants first rated the vignettes on a scale of 0 (worst imaginable health) to 100 (full health) on a visual analogue scale (VAS). Participants were asked to read the vignettes one at a time, including a state called ‘dead’ and place them on this scale. In the TTO task participants imagined that they were currently experiencing each health state (described in the vignette) and they were asked to choose whether they preferred: (1) to live in the health state for a period of 10 years followed by death; (2) to live for X number of years in full health; or (3) to indicate that the two previous options were equally desirable. Time in the state of full health (X) was systematically reduced from 10 downwards until the respondent was indifferent between the two choices.
Analysis
Descriptive statistics were used to present socio-demographic data such as age, gender, qualification and employment. The VAS and TTO data were analysed separately using regression modelling using SAS software version 9.4. In these analyses, the dependent variable was a transformation of the VAT and TTO values, to transpose left-skewed utility data into right skewed dependent variables so that distributions could be more easily fitted to the data. These transformations consisted in changing each value into its complement: TTO complement = (1- TTO utility) and VAS complement = (100- VAS value). In order to obtain the actual utility values for each health state, the opposite linear transformation needs to be applied: TTO utility = 1 - TTO complement, and VAS value = 100 - VAS complement.
The independent variable in each regression was the health state, in order to obtain a utility value with a 95% confidence interval for each HLH-related health state. Based on these regressions, the mean TTO and VAS values (and 95% confidence intervals) were generated for each health state. In a second phase, gender, age, employment and education were also included in the regression, to understand whether any of the respondent characteristics had an influence on the results.
The analysis was carried out in the General Estimating Equations framework, which is suitable for analysing correlated data. As each respondent evaluated the full set of health states, these evaluations are correlated within the subject. Several forms of the correlation matrix between the repeated measurements were fitted: independent, exchangeable, compound symmetry and unstructured correlation matrix. Furthermore, models were estimated with an identity or a log link, and with a normal or Gamma distribution for the error terms. These models were tested against each other using the quasi-likelihood under the independence model criterion (Quasi Information Criterion or QIC) which was developed by Pan for model selection in a GEE environment [9, 10]. This QIC statistic works in an analogous way to Akaike’s information criterion (AIC) in that the best fitting model is that with the lowest value of QIC after paying a penalty for the number of parameters fitted [11].