Health related quality of life in 3 and 4 year old children and their parents: preliminary findings about a new questionnaire

Our sample included all surviving babies admitted for more than 24 hours to one of 3 level III NICUs in British Columbia (Canada) over a 16-month period (March 1996 through June 1997 inclusive). These 3 units (at Royal Columbian Hospital, Victoria General Hospital and British Columbia Women's and Children's Hospital) provided 100% of the tertiary care NICU beds in the province at the time. Mothers' name and contact details were obtained from each hospital. This population of babies was then matched with provincial mortality records to identify and exclude any babies that had died after discharge from the NICU. To ensure the data were independent, only families with one child in the study sample were included in this paper.

Our comparison sample of healthy term babies was recruited from the two hospitals with an affiliated hospital-based primary care unit (BC Women's and Children's Hospital and the Royal Columbian Hospital). This sample included all babies delivered over 11 months (March 1996 through January 1997 inclusive) by any primary care physician from these two units working within either of these two hospitals. Multiple births, babies with a sibling in the NICU sample, and babies subsequently admitted to a NICU for more than 24 hours were excluded. Contact details for the mother were obtained from the health records department at one hospital and directly from the primary care unit at the other.

A questionnaire booklet, that included a number of separate instruments, was sent to each mother as her child turned 3 1/2 years of age. A consent letter was included to obtain permission to link the questionnaire data with hospital birth records. The caregiver that had, to that point in the child's life, spent the most amount of time with the child was asked to complete the questionnaire. Non-respondents were sent a reminder letter and up to two more copies of the questionnaire as necessary. Finally, phone calls were made as part of a final effort to reach families. If the telephone number was not in service or reassigned, or a questionnaire was returned to us from the post office as undeliverable, a comprehensive search strategy was implemented. The process involved searching the Internet and/or contacting the mothers' primary care physician to obtain an address.

The questionnaire booklet included the developmental full-length version of the Infant Toddler Quality of Life Questionnaire (ITQOL) [9, 29]. The prototype contains 103-items that measure 8 infant and 5 parental concepts (see Table 1). This instrument was developed for infants as young as 2 months and toddlers up to five years of age using developmental guidelines used by pediatricians and other published literature [12]. More than half the items in each scale must be answered in order to derive a score. Raw scores are calculated for each scale by computing the algebraic mean of the items. Following published convention [30], raw scores are then transformed to a scale from 0 (worst health) to 100 (best health).

Data completeness was measured by computing the percentage of items completed for each scale and the instrument. Following published conventions [31‐36], item-to-scale correlations (corrected for overlap) were considered satisfactory for items that correlated .40 or more with their hypothesized scale. Item discriminant validity was considered successful if the correlation between an item and its hypothesized scale was significantly higher (≥ 2 S.E.) than correlations between that item and all other scales. As advised with newly created scales [30], the percentage of correlations that were ≥1 S.E. higher for each item and its hypothesized scale were also examined.

For each scale, we determined the percentage of scores that could be computed. The distribution of scores was examined to determine potential floor and ceiling effect (i.e., people scoring at the absolute lowest and highest ends of the continuum for each scale). Scale internal consistency was assessed in terms of Cronbach's α coefficient. Internal consistency was considered satisfactory if the coefficient was at least .70 [37, 38]. To evaluate the degree to which each scale was "unique", correlations among all scales were examined and compared against the respective Cronbach's α reliability coefficient observed for each individual scale. In general, the correlation between scales should be less than the alpha coefficient achieved for an individual scale [37]. To examine test-retest reliability, a random sample of 80 NICU respondents, who indicated they would be willing to participate in further research, was contacted by telephone. Those that agreed to participate were sent a copy of the ITQOL in the mail. A second copy of the questionnaire was mailed out once it was confirmed that the first copy had been completed. Test-retest reliability was assessed through intra-class correlation coefficients. ICCs of at least .70 were considered satisfactory [37, 38]

To test concurrent validity, scale scores in the ITQOL were correlated with scores for similar and dissimilar scales in three validated instruments: the Child Behavior Checklist/1.5-5 (CBCL/1.5-5)[39]; the SF-36 [40, 41]; and the Family Assessment Device (FAD) [42]. Scales from each instrument that are intended to measure similar constructs should have higher correlations (convergent validity) with each other than with scales that measure unrelated constructs (divergent validity). Correlations of <0.20 were considered negligible; 0.20 to 0.34 weak; 0.35 to 0.50 moderate; and >0.50 strong [43].

Since no validated multidimensional generic measure of HRQL was available for validation purposes, we used a measure of behavior as 55% of items in the ITQOL measure child behavior or temperament. The CBCL/1.5-5 measures behavioral, emotional and social functioning in children 1 1/2 to 5 years of age. This 100-item instrument measures both internalizing and externalizing syndromes and can be summed to produce a total problem score. A higher score reflects greater presence and severity of symptoms.

The SF-36 [40, 41] assesses the following 8 domains of adult health: physical health; physical role limitations; emotional role limitations; mental health; social function; energy; pain; and general health perception, and was used to help validate the ITQOL parent-impact scales. Since the mental health domain and one item from general health perception are included in the ITQOL, the remaining 6 domains were used in the validation process. Scores on these domains can range from 0 (worst health) to 100 (best health).

The Family Assessment Device (FAD) [42] is a measure of family functioning and was used to help validate the Family Cohesion item. Scores for this 12-item scale can range from 0 to 36 with higher scores indicating greater dysfunction.

The ability of the ITQOL to discriminate between groups of children with poorer expected outcomes was determined by comparing ITQOL scale scores for the following two dichotomous variables (using Mann-Whitney U-test for statistical significance): (1) NICU vs. healthy baby sample; and (2) children with one or more health problems (from a list of 16 common childhood conditions) vs. children with no health problems. The NICU sample and the group with one or more health problems were expected to have poorer reported health. Effect size statistics (i.e., mean difference divided by pooled s.d.) were computed to determine the magnitude of the difference in mean scores.

Item-level results are presented in Table 2. Sixty-seven percent of respondents in the NICU sample and 74% of respondents in the healthy baby sample answered all 103 items. This was lower than the 83% (NICU sample) and 94% (healthy baby sample) of respondents who completed all items for the similar length CBCL/1.5-5. Of those missing at least one response on the ITQOL, three quarters of respondents in both samples missed answering only 3 items or less. The rate of missing data within each scale varied from 2.3% (Impact-emotional) to 8.9% (General Health Perception) for the NICU sample, and from 0.8% (Impact-emotional) to 7.8% (Temperament and Moods) for the healthy baby sample.

For item-scale correlations, 91% (NICU sample) and 84% (healthy baby sample) of items correlated with their own domain above the recommended standard (0.40). Within domains, perfect results were obtained for 7 (NICU sample) and 5 (healthy baby sample) scales. For item-discriminant validity, 97% (NICU sample) and 87% (healthy baby sample) of items correlated more highly (≥ 2 S.E.) with their hypothesized scale than with other scales. Perfect results (100%) were attained for 8 of the 10 scales in the NICU sample, and 6 of the 10 scales in the healthy baby sample. Only 2 items in the NICU sample (in Getting Along) and 5 in the healthy baby sample (in Temperament and Moods, General Behavior and Getting Along) did not correlate ≥ 1 S.E. with its hypothesized scales.

Scale-level results are presented in Tables 3 and 4. The proportion of missing values for scored domains was small: 2.9% (Physical Abilities) or less. There were no floor effects, but ceiling effects (scores of 100%) were apparent. The largest ceiling effect (69.3% NICU; 85.8% healthy baby) was in the Physical Abilities scale. The range of scores was particularly skewed for three scales (Physical Abilities, Growth/Development, Bodily Pain) where more than 84% of respondents in both samples reported scores of 75 or higher. Scores for scales that assess aspects of emotional and behavioral function showed more variability.

For both samples, the Cronbach's alpha coefficients were .80 or higher. One scale (Physical Abilities) achieved a coefficient of .96. The correlations between the ITQOL scales were on average moderate (see Table 5 and 6). All correlations between scales were less than their reliability coefficients, providing evidence of unique reliable variable measured by each scale.

For test-retest reliability, 2 copies of the ITQOL were received from 71% of the families who agreed to participate. Mailings were separated on average by 13 (s.d. 5) days. Intra-class correlation coefficients exceeded the .70 benchmark, and were as follows: Physical Abilities = .80; Growth/development = .84; Bodily Pain = .71; Temperament = .75; General Behavior = .94; Getting Along = .87; General Health Perception = .89; Mental Health = .83; Impact-emotional = .77; and Impact-time = .75.

Correlations between related scales in the ITQOL and other standardized instruments were strong (see Tables 7 and 8). Specifically, Getting Along, Temperament, and General Behavior correlated more strongly with CBCL syndrome and total problem scores and less strongly with domains that measure aspects of physical health. Similarly, as anticipated, the parental impact scales (emotional and time) correlated more strongly with SF-36 psychosocial scales than with SF-36 physical scales. The family cohesion item correlated strongly with the Family Function Scale and weakly or moderately with all other scales.

Table 9 and 10 presents findings for tests of discriminant validity. Parents of NICU children reported their children as having significantly poorer HRQL than children in the healthy baby group for 5 of the child scales. Scores for the NICU sample were also lower for the 3 parent scales. These differences were all small in size (effect size .44 or smaller).

In the NICU sample, those with children with at least one health problem that required treatment in the past year had poorer reported HRQL in all areas compared with those without health problems. In the healthy baby sample, significant differences were noted for 4 of the child scales.