Shortening the paediatric Haemophilia Activities List (pedHAL) based on pooled data from international studies

2.1 Study design and study population

This study was a cross-sectional multicentre secondary analysis of pooled data of published studies using the pedHAL. The Medical Research Ethical Committee (MREC) of the University Medical Center Utrecht reviewed the study (protocol number 18-309/C).

2.2 Pooling of published pedHAL data

A literature search identified five studies published between 2010 and April 2018, which used the pedHAL in children with haemophilia. In addition, two studies in preparation for publication were identified and included. Inclusion criteria were children with haemophilia A (FVIII) and B (FIX) of all severities, aged 4–18 years. Patients were excluded if more than half of the pedHAL items were missing. If both the children and parent proxy pedHAL scores were available, only the pedHAL completed by the child was included in the analyses. Authors of all seven papers were invited to share the original pedHAL data (scores per item) and de-identified patient characteristics. All but one authors accepted our invitation. Only data of children with haemophilia A (FVIII) and B (FIX) aged 4–18 years who participated in the selected studies were included.^{1, 4, 7, 9-11}

2.3 Measurements

The pedHAL assesses self-reported limitations in activities and participation in children with haemophilia. It consists of a patient version (8–18 years) and parent version (4–18 years) both with 53 items, distributed over seven domains. Patients score the items on a 6-point Likert scale (‘impossible’, ‘always’, ‘usually’, ‘sometimes’, ‘almost never’, ‘never’), with a ‘not applicable (N/A)’ scoring option. Domain scores and sum scores are converted to a normalized domain score ranging from 0 (worst possible functional abilities) to 100 (best possible functional abilities) in the scoring tool (available at www.vancreveldkliniek.nl). According to the pedHAL scoring manual, domain scores were only calculated if half or more of the items of a domain were scored on the 6-point Likert scale.

Patient characteristics analysed included age at pedHAL assessment, type of haemophilia (A or B), severity of the disease (mild [factor VIII/IX activity 0.06 – 0.40 IU/ml], moderate [factor VIII/IX activity 0.01 – 0.05 IU/ml] or severe [factor VIII/IX activity <0.01 IU/ml]), clotting factor regimens (prophylaxis yes/no and start prophylaxis before age of 3 years yes/no) and current inhibitor status.

2.4 Statistical analyses

Patient characteristics were presented as proportions or medians (interquartile ranges [IQR:P25 - P75]). A Kruskal-Wallis test was performed to compare age according to treatment regimen (prophylaxis start <3 years vs. prophylaxis start ≥3 years vs. no prophylaxis). Descriptive analyses (median, IQR, range, mean and standard deviation [SD]) were performed for the pedHAL domain and sum scores. Based on reported limits of agreement (LoA) of test-retest data,¹ limitations in activities and participation were defined as ≤95 points for domain and sum scores.

Non-informative items were identified in a stepwise process (7 steps) according to the method of de Vet et al. (2011), from the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) initiative.¹² After each step, non-informative items were deleted, before proceeding with the following step.

Step 1: Per item the proportions of each scoring option (‘impossible’, ‘always’, ‘usually’, ‘sometimes’, ‘almost never’ or ‘never’) was determined, excluding the missing and ‘N/A’ scored questions. Proportions of minimum (‘impossible’) and maximum (‘never’ problems) scores were analysed to detect floor and ceiling effects. Items with ≥85% minimum or maximum scores were removed.

Step 2: The missing data and scores with ‘N/A’ were examined. The authors removed items which were scored >15% as ‘missing’ or ‘N/A’.

Step 3: Inter-item correlations were evaluated. Inter-item correlations calculated with Spearman's rho <0.2 indicated items which do not correlate with any of the others and >0.9 indicated item redundancy. Items with inter-item correlations <0.2 and >0.9 were not included in the factor analysis.

Step 4: Component loadings on exploratory factor analysis were evaluated. Items were analysed on categorical level. Items with factor loadings <0.5 were removed. Model fit was evaluated with the Root Mean Square Error of Approximation (RMSEA); <0.08 indicates moderate model fit and <0.05 indicates good model fit.

Step 5: Inter-item correlations were evaluated for the second time. Inter-item correlations calculated with Spearman's rho >0.7 within one factor were indicators for item redundancy. Items which had a correlation >0.7 were reviewed by IK, KF and JJ and one of the items was removed.

Step 6: Internal consistency calculated with Cronbach's α and internal consistency after item deletion were compared. Cronbach's α should be 0.7 and 0.9; a higher Cronbach's α after item deletion was considered a reason to eliminate an item.

Step 7: Item-total correlations for the pedHAL total score were evaluated. Item-total correlations were calculated with Spearman's rho. Items with item-total correlations <0.3 were removed.

A sensitivity analysis was performed where only parent scores were analysed in cases where both were available (n = 72).

After removing non-informative items, a pedHAL_short was created. Median (IQR) normalized sum and domain scores and percentages of scores ≤95 points were calculated for the pedHAL_short, similar to the calculation of sum and domain scores in the original scoring tool. The differences between the pedHAL and pedHAL_short sum scores were calculated and shown in a box plot. In addition, a one-way analysis of variance (ANOVA) was performed to compare differences according to treatment regimen (prophylaxis start <3 years vs. prophylaxis start ≥3 years vs. no prophylaxis). A secondary exploratory factor analysis was performed for the pedHAL_short to detect possible underlying constructs.

SPSS (version 25, IBM) was used for data analyses. Mplus (version 6.12, Muthen & Muthen) was used for the exploratory factor analysis.

3.1 Patient characteristics

From the data of six studies, 315 children with haemophilia A or B were included. The data are from the Netherlands (n = 84),^{1, 9} Romania (n = 28),⁴ United Kingdom (UK) (n = 123),⁷ Lithuania (n = 15) ¹⁰ and Canada (n = 65).¹¹ Patient characteristics are shown in Table 1. Median age at the time of completing the last pedHAL was 12.2 years (range 4.0 – 18.0) and was similar for patients with different treatment regimens (P = 0.22). The majority of the patients had severe haemophilia (87.3%). One patient was excluded because he completed less than half of the items of the questionnaire. The bulk of the questionnaires (81.3%) that were analysed were completed by the children, and the others were completed by parents.

3.2 PedHAL domain and sum scores

Domain and sum scores are shown in Table 2. The median (IQR) pedHAL sum score was 96.7 (88.0 – 100.0). ‘Positive’ pedHAL sum scores (≤95 points) were observed in 43% of participants. The median (IQR) domain scores were lowest for the domains ‘sitting/kneeling/standing’ (97.8 [82.2 – 100.0]), ‘functions of the legs’ (97.8 [83.6 – 100.0]) and ‘leisure activities and sports’ (97.8 [80.0 – 100.0]). The other domains had median scores of 100.0. Domain scores were not calculated for 2 participants for the domains ‘functions of the arms’ and ‘self-care’, and for up to 60 participants for the domain ‘leisure activities and sports’, because more than half of the items were scored as missing or ‘N/A’.

3.3 Item reduction

The stepwise process to select non-informative items is shown in Table 3. The frequency tables generated for step 1 and 2 and the table with item-total correlations for step 7 in the item reduction process are shown in the Data S1.

1. Floor and ceiling effects.

Minimum and maximum scores were evaluated for all items. There was no floor effect in any pedHAL item. Ceiling effects were shown in 2/6 items of the domain ‘functions of the arms’ and in 6/9 items of the domain ‘self-care’.

2. Missing data and scores with ‘N/A’.

There were small numbers of missing responses (0–3) on the items. Missing and/or ‘N/A’ responses were scored in >15% of the children in 1/3 items of the domain ‘use of transport’, 2/3 items of the domain ‘household tasks’ and in 9/11 items of the domain ‘leisure activities and sports’.

3. Inter-item correlations (1).

None of the items had correlations with other items lower than 0.2 or higher than 0.9. All remaining items were used for the exploratory factor analysis.

4. Component loadings of the exploratory factor analysis.

Table 4 shows the component loadings of the exploratory factor analysis. The exploratory factor analysis suggested no items were eligible for item reduction. A 2-factor model was selected which included all remaining items, and the two factors were identified as arm activities and leg activities. The model fit of the 2-factor model was 0.07 (RMSEA), indicating moderate model fit. The factor loadings were >0.5.

5. Inter-item correlations (2).

Inter-item correlations were re-evaluated. In the domain ‘sitting/kneeling/standing’, 3/10 items, which had inter-item correlations >0.7 with other items, were removed. In the domain ‘functions of the legs’, 5/11 items were removed. The items ‘running’ and ‘jumping’ had a correlation of 0.73. The authors decided to remove the item ‘jumping’, which was considered less relevant in lifelong outcome assessment. The items ‘walking upstairs’ and ‘walking downstairs’ had a correlation of 0.81. The authors decided to remove the item ‘walking downstairs’, which was scored as less difficult than ‘walking upstairs’ by the participants. In the domain ‘use of transport’, the item ‘using public transport’ (1/3) was removed. In the domain ‘self-care’, 2/9 items were removed.

6. Internal consistency calculated with Cronbach's α.

The remaining 22 PedHAL items were strongly related (Cronbach's α of 0.97), which indicates redundancy of items. Only complete cases (n = 201, 63.8%) were included in the analysis. The Cronbach's α after deletion of separate items was equal or smaller, which did not identify candidate items for removal. Eventually, the authors decided to keep the remaining 22 items, because the Cronbach's alpha was already lowered by removing the 31 items.

7. Item-total correlations for pedHAL total scores.

All item-total correlations were high (Spearman's rho =0.55 – 0.76), thus identifying no candidates for item reduction.

The sensitivity analysis with parent proxy (n = 131) and child (n = 184) forms resulted in a shorter pedHAL_short (20 items). The items ‘walking or riding up a small hill or slope without help’, ‘stretching to reach something above your head’ and ‘putting on pants’ were removed in step 5 (inter-item correlations), and ‘putting on shoes and socks’ was not removed in step 5.

3.4 PedHAL_short with 22 items

In Table 3, all items of the pedHAL_short are shown. Domain and sum scores of the pedHAL and pedHAL_short are shown in Table 5.

Twenty-two items remained after removing the items (n = 31) according to the seven steps. All domains were still represented in the pedHAL_short. Most items of the pedHAL_short belonged to the domains ‘sitting/kneeling/standing’ (n = 7) and ‘functions of the legs’ (n = 6). For the domains ‘use of transport’, ‘self-care’ and ‘household tasks’, only one item remained in the pedHAL_short. The median (IQR) pedHAL_short sum score was 97.3 (87.0 – 100.0), which was similar to the pedHAL sum score. The differences between the pedHAL and pedHAL_short sum scores were similar in patients receiving prophylaxis started <3 years and started ≥3 years and patients receiving no prophylaxis (P = 0.82) (see Figure 1). The domains had median scores of 100.0, in exception of a median domain score of 97.1 for ‘sitting/kneeling/standing’. Domain scores for ‘sitting/kneeling/standing’, ‘functions of the legs’, ‘use of transport’ and ‘self-care’ were higher than the original pedHAL domain scores. Domain score for ‘functions of the arms’ was lower than the original pedHAL domain score. The largest discrepancy in the proportions of abnormal domain scores (≤95) was observed for the domain ‘use of transport’ (pedHAL: 26% vs. pedHAL_short: 16%), which was a result of removing the item ‘cycling’.

The secondary exploratory factor analysis with the 22-item pedHAL_short resulted in a 1-factor model, indicating that a sum score containing all 22 items needs to be used.

4.1 Internal and external validity

This is the first study presenting pedHAL data of 315 patients from heterogeneous populations. In these published pedHAL data, the majority of patients had received prophylactic treatment (80%) and half of the patients had early prophylaxis (51%). Ceiling effects are more likely to occur in intensively treated patients.

Despite the lack of cross-cultural validation studies of the pedHAL, the use of the pedHAL is recommended in international guidelines.¹³ After development of the pedHAL in Dutch children, only one clinimetric study was performed in Romanian children.^{1, 4} In Romanian children, high proportions of ‘N/A’ responses were recorded in the domains ‘household tasks’ and ‘leisure activities and sports’. This was confirmed by the present ‘pooling’ study including more children from different populations. It seems that especially these domains are culturally dependent and removing these items is expected to results in a questionnaire that will perform better in a multicultural and global context. The shift towards higher domain scores in some domains (ie ‘sitting/kneeling/standing’, ‘functions of the legs’, ‘use of transport’ and ‘self-care’) and lower domain scores for ‘functions of the arms’ were a result of the different reasons for removing items. For example, in the domain ‘functions of the arms’, items were only removed for ceiling effects and in the domain ‘use of transport’ the most difficult item ‘cycling’ was removed for a high number of ‘N/A’ responses. However, the sum scores of the pedHAL and pedHAL_short were similar.

For two items with a high item-total correlation, rephrasing of the question may be considered. The items ‘walking upstairs’ and ‘walking downstairs’ had a high inter-item correlation of 0.81. ‘Walking upstairs’ was reported by the participants as being slightly more difficult. As both items are about walking stairs, ‘walking stairs’ may better capture the activity than choosing one of the two activities. For calculating the pedHAL_short from the original pedHAL, any limitation reported on walking stairs could be scored as abnormal.

Internal consistency of the pedHAL_short (Cronbach's α = 0.97) is still higher than the recommended Cronbach's α between 0.7 and 0.9. As the internal consistency improved after reduction of the 31 items and there was no clear indication for removing any other specific items, it was decided to retain the remaining items.

To ensure that the pedHAL_short contains all informative items, the 22-item pedHAL_short was preferred above the 20-item pedHAL_short resulting from the sensitivity analysis including all parent forms.

4.2 Comparison with other studies

Similar domains were important in two studies not included in this pooled data. In boys with haemophilia from Lithuania and Portugal, the most difficulties were reported in the domains ‘sitting/kneeling/standing’, ‘functions of the legs’ and ‘leisure activities and sports’.^{8, 14} The fewest difficulties were reported in the domain ‘self-care’ in the Lithuanian data and in the domains ‘household tasks’ and ‘self-care’ in the Portuguese data, which were both less informative domains in the pooled data.^{8, 14} Exact scores were difficult to compare because both studies reported mean scores. The highest proportions of ‘N/A’ responses were in the domain ‘leisure activities and sports’ in the Lithuanian data, which was similar in the pooled data.⁸ No other studies were available to further compare our findings. In adults, similar scoring patterns were shown by domain level.^{15, 16}

4.3 Clinical implications and future research

Within a context of multiple outcomes assessments in haemophilia care, a shorter assessment of limitations in activities and participation is desirable. This pooling study of international pedHAL data in children with mild to severe haemophilia with a wide range of treatment regimens suggested that 31 pedHAL items are redundant, resulting in a notable shortening of the questionnaire. The shorter version of the pedHAL includes the most relevant and informative items for children and youth with haemophilia. The pedHAL_short can be derived from the original pedHAL, which allows for use in longitudinal studies. Only the sum score should be used for the pedHAL_short, since some domains only have 1 item in the pedHAL_short. Before introduction of the pedHAL_short construct, validity and reliability of the questionnaire should be investigated in diverse populations.