The impact of factor infusion frequency on health-related quality of life in people with haemophilia

The ‘Cost of Haemophilia in Europe: a Socioeconomic Survey’ (CHESS) study was a cross-sectional, retrospective study carried out in 2015, where patients aged ≥18 years with severe haemophilia in five European countries (France, Germany, Italy, Spain and the UK) were invited to participate ^[14]. 1,285 patients were recruited by 139 haematologists and haemophilia care providers based in hospitals and clinics. Data were collected using two questionnaire forms: the patient record form (PRF) was completed by physicians and a Patient and Public Involvement and Engagement form (PPIE) was completed by individual patients. Full details of the methodology for data collection have been published ^[14].

Physicians reported the number of comorbidities diagnosed at the time of consultation, hospital admissions in the previous year, the presence of a target joint (binary variable), level of adherence to prescribed factor replacement therapy (high or low/medium, based on the physician's records) and the severity of chronic pain (according to clinical assessment). In CHESS, primary prophylaxis was defined as receiving prophylaxis since the start of treatment for haemophilia; secondary prophylaxis was defined as switching to prophylaxis from on-demand treatment ^[15]. We included only patients taking a prophylaxis dose of ≥10 IU/kg, which is the minimum effective dose ^[16]. Patients currently receiving on-demand treatment were not included due to their infrequent use and different requirements for adherence. Patients with a history of inhibitors or currently with inhibitors were also excluded.

Additionally, patients were excluded when the ratio of factor dose to body weight exceeded 1.2 (calculated as FVIII: IU per infusion/body weight x 2; FIX: IU per infusion/body weight) because this suggested the possibility of an increased risk of thrombosis or of error in completing the questionnaire.

Joint bleeds were defined as episodes of bleeding into joints diagnosed by the physician which were recorded as major and minor bleeds. Minor bleeds were defined as those with mild pain, minimal swelling, minimal restrictions of motion and resolution within 24 hours of treatment. Major bleeds were defined as pain, effusion, limitation of motion and failure to respond to treatment within 24 hours. The presence and number of target joints was defined as joints affected by chronic synovitis diagnosed by the physician. This definition, originally proposed in 2004 ^[17], was adopted following discussion with an expert advisory group.

All participants in the CHESS study provided informed consent. The CHESS protocol was approved by the Research Ethics Sub Committee of the Faculty of Health and Social Care within the University of Chester. The approval stipulated that the study was to be carried out in correspondence with regional and relevant guidelines.

HRQoL was assessed using the validated tool EQ-5D-3L ^[18,19]. This involves patient self-reporting of their health status in five dimensions: mobility, self-care, usual activities, pain and discomfort, anxiety and depression. EQ-5D-3L index values are normally confined to the range −0.594 to 1.0, where 1.0 represents ‘perfect health’, zero represents ‘dead’, and values less than zero are health states ‘worse than dead’. We were unable to utilise predictive methodologies confined to a 0–1 range (e.g. logistic regression, beta regression) without transforming the data EQ-5D-3L values in the sample at zero. The distribution of EQ-5D-3L values was therefore shifted so that, in this analysis, EQ-5D-3L values ranges from zero to 1.594 (as zero meaning worse than dead and 1.594 perfect health). Shifting EQ-5D-3L values does not change their distribution and will facilitate the use of a more appropriated statistical method given the complex distribution. Other techniques have been proposed, such as adding a threshold in the upper level of the EQ-5D-3L scale range (Tobit models), but the values obtained from this study are very close to the threshold which would bias the results ^[20].

Descriptive analysis was conducted to summarise patient characteristics. Means were used to describe continuous variables and frequency and proportions to describe categorical variables. Independent sample t-tests and chi squared for independence tests were conducted to test for between-group differences.

Multivariate analysis was carried out using generalised linear regression models. The analyses were performed using R software and the function glm(). Three models were generated with a single set of candidate covariates: the first examined the effect of the covariates on quality of life, the others examined their effect on minor and major bleeds. A gamma distribution with log link function was assumed for a quality of life model whereas a Poisson with log link function was used for the major and minor bleeds regression models. For each model, a selection approach was applied to identify the best model given its complexity and goodness of fit. Candidate covariates included all potentially important determinants of the outcomes. The model selection was based on a top-down selection, which consists of having a model with all candidate covariates and removing each covariate sequentially to see the improvement of the fit. Likelihood ratio test was the test used in the model selection process.

The set of candidate covariates were:

y = age + target joint + subtype + treatment strategy + number of hospital admissions + number of comorbidities + chronic pain + adherence + infusion +(infusion rate * chronic pain)

Patient characteristics are summarised in Table 1. From the CHESS population (n=1,285), excluding 58 with a current inhibitor, of 1,227 patients with severe haemophilia, 338 (26%) provided responses for major and minor bleeds and target joints and 145 (11%) provided an EQ-5D-3L response. Most patients had haemophilia A. The other participants were not included in the analysis as described in the materials and methods section.

The majority of participants were young adults (<30 years old) and the mean level of comorbidity was correspondingly low, though there was marked variation in the number of concomitant conditions; this was matched by variation in admission frequency. Prophylaxis was secondary in almost two thirds of patients; the number of infusions weekly ranged from 1–7. Physicians reported that almost two thirds had a rating of low adherence (on a Likert scale of 1–2 vs 3–4) to their prophylaxis regimen. Physicians reported pain in 64% of patients. This was rated mild in most cases but considered to be moderate in one in five patients, with few having severe pain.

Compared with the CHESS population as a whole, the subset included in this analysis were slightly younger, more likely to be using primary prophylaxis, and fewer had high adherence. However, both had more people on three times weekly administration and the distribution of pain severity was similar.

The coefficients from the generalised models were exponentiated (exp{β}) to represent the proportional change in the response variable (EQ-5D-3L score, number of major and minor bleeds). The generalised model for major bleeds showed that using primary (rather than secondary) prophylaxis, number of comorbidities, infusion rate and having mild or severe chronic pain were not statistically significant variables (Table 2). Similarly, the interaction between infusion rate and pain was not significant. In the model of best fit, major bleeds were positively associated with age (i.e. higher number of bleeds were more frequent with higher age), having a target joint and hospital admissions whereas high prophylaxis adherence was associated with less events of major bleeds. Pain at every level of severity was positively associated with number of major bleeds.

In the case of minor bleeds, the generalised model also showed no significant association with primary prophylaxis vs secondary prophylaxis, comorbidities, mild pain, the interactions of infusion rate and pain, and also with age or haemophilia subtype. In the model of best fit, the presence of target joints and admissions were positively associated with minor bleeds whereas high adherence and higher infusion rate were associated with fewer minor bleed events. The association was also positive for chronic moderate and severe pain, but not mild pain, and for the interaction between infusion rate and severe pain.

There was no significant association in the generalised model between HRQoL and age, target joints, comorbidities, haemophilia subtype, adherence, infusion rate, or mild or moderate pain. The model of best fit showed that HRQoL was positively associated with using lifelong prophylaxis but negatively associated with number of hospital admissions. Patients with severe chronic pain reported a substantial negative impact on HRQoL; this was slightly but significantly improved by increases in the annual infusion rate.

Analysis of the impact of infusion rate on HRQoL shows a trend to incremental decrease in EQ-5D-3L score with each additional infusion per week (equivalent to 52 per year) by approximately 2%, increasing to 10% reduction with daily infusions.