Developing a non-categorical measure of child health using administrative data

by Rubab G. Arim, Dafna E. Kohen, Jamie C. Brehaut, Anne Guèvremont, Rochelle E. Garner, Anton R. Miller, Kimberlyn McGrail, Marni Brownell, Lucy M. Lach and Peter L. Rosenbaum

The increasing availability of provincial and national linked data has created opportunities to conduct health research using administrative information. Studies of children that used such data have largely focused on health service utilization or on the prevalence of specific conditions.Note 1-3 However, categorical approaches based on specific conditions do not include the large numbers of rare conditions that have similar consequences, nor do they include children with undiagnosed conditions. Moreover, policies, programs, and interventions generally do not target specific conditions.

An alternative conceptual framework offers a non-condition-specific or “non-categorical” approach to identifying children with chronic illness.Note 4 In this framework, children can be grouped together by virtue of similar consequences of diverse conditions.Note 5 This non-condition-specific approach is particularly useful for program planning, service delivery, and policy development,Note 6-8 where decisions must be made on the basis of health consequences for large groups or populations.

Neff et al.Note 9-11 used medical claims and hospital discharge data to study children with chronic conditions over time. A unique feature of their research is that, despite the use of a categorical or specific clinical condition as input, the resulting conceptualization of children with health care needs incorporates medical complexity with a focus on consequences (non-condition-specific). Their findings suggest that ,although administrative data typically pertain to services and diagnoses (condition-specific), research using non-condition-specific approaches may be possible.

The present study examines the feasibility and limitations of applying a non-categorical approach (focused on service utilization rather than on specific diagnoses) to administrative data in order to identify children with health problems. The widely used Children with Special Health Care Needs (CSHCN) Screener is one of the few established measures of child health that employs a non-categorical approach. The Screener is a validated, non-condition-specific, consequence-based tool to identify children with health problemsNote 5,Note 12,Note 13,Note 14 as reflected in the use of health services, devices, and medications. However, the Screener is survey-based and has not been applied to administrative data. With information made available by Population Data British Columbia, this article explores the opportunities and limitations of applying the CSHCN Screener to linked administrative data.

Data and methods

Data

The analysis is based on four separate administrative data holdings of Population Data British Columbia (PopData):

  • The Medical Services Plan (MSP) Payment Information File: medical services provided by fee-for-service practitioners to people covered by British Columbia’s universal health insurance program from 1985 to the present.Note 15,Note 16
  • The Discharge Abstract Database (DAD)Note 17: hospital discharges, transfers, and in-hospital deaths of patients from all acute-care hospitals from 1985 to the present.
  • PharmaNetNote 18: prescription data for drugs and medical supplies (for example, insulin pumps, orthotics) dispensed by pharmacies.
  • The Consolidation File (MSP Registration and Premium Billing): population demographic data prepared for research use by PopData.

All PopData data are linked using encrypted randomly assigned identifiers that match individual personal health numbers, and thereby provide individual-level data that maintain anonymity and confidentiality.

This analysis is part of a longitudinal study of the health of caregivers of children with health problems. The sample of children (N = 232,870) was based on the most recent data available when access was requested (March, 2008). This study imposed the CSHCN Screener criterion that a health condition must have lasted or be expected to last for at least 12 months. The analysis used 2006 data (January 1 to December 31) for children aged 6 to 10 (the age range of children in the larger study) and incorporated data for calendar years 2005 and 2007 to capture conditions that lasted or were expected to last for at least 12 months.

Data access

The study was approved by Research Liaison staff of PopData, the Data Stewards at the British Columbia Ministry of Health, and the Ottawa Health Science Network Research Ethics Board, where the research was carried out. The encrypted data files were made accessible to the research team on a secure research environment (SRE) through PopData. Analysts accessed the SRE at the Ottawa Hospital Research Institute, an affiliated institute of the University of Ottawa.

Applying the CSHCN Screener

The CSHCN Screener toolNote 12 is a parent-reported measure used to identify children with special health care needs; it has been validated with survey data.Note 13,Note 19-21 Five health consequences, which are not mutually exclusive, are assessed: 1) need for or use of medicine prescribed by a doctor (other than vitamins); 2) need for or use of more medical care, mental health care, or educational services than is usual for children of the same age; 3) being limited or unable to do the things most children of the same age can do; 4) need for or receipt of special therapy, such as physical, occupational, or speech therapy; and 5) emotional, developmental, or behavioural problem for which treatment or counselling is needed or received. A “yes” to any item (except 5) triggers two follow-up questions: whether the consequence is attributable to a medical, behavioural, or other health condition, and whether the condition has lasted or is expected to last for at least 12 months. The fifth item triggers only the latter question about duration.

Appendix Table A explains how each CSHCN Screener item was identified in the relevant administrative database. Only item 3 identified children based on specific diagnostic codes. The items were identified in the MSP administrative data by ensuring that 1) the specific health consequence was coded more than once (for example, two similar diagnostic codes), and 2) at least one of these codes appeared within the target 12-month period (with the second appearing within 6 to 18 months before or after the first code), which required inclusion of data from the previous (2005) and subsequent year (2007). If a specific health consequence was identified in the administrative data within the specified time frame, children were included in the CSHCN group; all others were grouped as non-CSHCN. Although the criteria are not mutually exclusive, this is in line with the original CSCHN Screener tool criteria. For example, although the same child can be captured by each of the five items, the final CSHCN group consisted of children captured by any one of the five items, thereby preventing double-counting.

Other measures for construct validation

Several individual- and community-level variables were used as construct checks on the coding of the five CSHCN Screener items. Because the prevalence of health problems is higher among boys than girlsNote 22,Note 23 and among older than younger children,Note 13,Note 23 child sex and age, as indicated in the Consolidation File, were examined. It was predicted that the CSHCN Screener items would be more prevalent among boys and among older children. Children classified as having health problems tend to be hospitalized more often than healthy childrenNote 22; therefore, overnight hospital admission, as indicated in the DAD, of the CSHCN and non-CSHCN groups was compared. Overnight hospital admissions in the DAD made it possible to access a data source other than the MSP to conduct construct checks on the coding of the CSHCN Screener items. Finally, socio-economic disadvantage has been associated with an increased risk of health problems.Note 13,Note 24 Consequently, it was predicted that the receipt of premium subsidies and residence in lower-income neighbourhood quintiles, as indicated in the MSP and Consolidation File, respectively, would be higher in the CSHCN group, compared with the non-CSHCN group. Neighbourhood income quintiles are an aggregated socio-economic status indicator that divides the provincial population into five income groups based on postal codes. PCCF+ softwareNote 25 was used to assign census geographies and income quintiles to MSP contract-holders.

Data analysis

Descriptive statistics were used to present the number and percentages of children identified for each CSHCN health consequence, including the three most common Anatomical Therapeutic Chemical (ATC) Classification groups that accounted for the highest percentages of prescribed drugs, medicines for which a prescription was filled, and recorded medical care service codes during the 2005-to-2007 period. Item prevalence was stratified by sex and age group. As an initial test of construct validity, rates of overnight hospital admissions, receipt of premium subsidies based on family income, and neighbourhood income quintiles were compared for children classified as CSHCN and those classified as non-CSHCN.

Results

Overall 17.5% (n = 40,840) of children were identified as meeting at least one of the five CSHCN Screener criteria; the majority―82.5% (n = 192,030)―did not meet any of the criteria (Table 1).

For item 1 of the CSHCN Screener (prescription medicine use), fewer than 1% of children (n = 1,602) were identified based on a 12-month cut-off (medicine use for 365 days of any 365-day period from 2005 to 2007). Based on a 9-month cut-off (medicine use 274 days or more), 3.3% of children (n = 7,596) were identified, and based on a 6-month cut-off (medicine use for 183 days or more), 5.6% of children (n = 13,036) were identified. The 9-month cut-off was used in this study (Table 1). The 6-month cut-off captured many medications used for acute rather than chronic conditions (for example, antiseptics and antibiotics), whereas the 12-month cut-off was considered too conservative, as it captured few respondents. To reinforce this decision, percentages of child hospital admissions by each medicine use cut-off were calculated—the percentages hospitalized were 6.5%, 8.4%, and 15.9% for the 6-, 9-, and 12-month cut-offs, respectively.

For children with 9 months of medicine use in any 12-month period from January 2005 to December 2007, the ATC groups that accounted for the highest percentages of prescribed drugs were: nervous system (48.5%; for example, psychostimulants), respiratory system (17.9%; for example, bronchodilators), and anti-infectives for systemic use (9.3%; for example, antibiotics). Methylphenidate (17.2%), Risperidone (6.1%), and Dexamfetamine (5.9%) were the top three medicines for which a prescription was filled.

For item 2 (medical care), the 95th percentile of the number of visits in 2006 was 14 for 6-year olds, 13 for 7- to 9-year-olds, and 12 for 10-year-olds. Based on these thresholds, 14% (n = 32,605) of children were identified as receiving more medical care than is usual for children the same age (Table 1). For children who received more medical care than usual, the three most commonly recorded service codes during the 2005-to-2007 period were regional examinations (52.2%), pathology (10.9%), and consultations (6.6%).

Based on the criteria for item 3 (functional limitations), 5.2% of children (n = 12,033) were identified. For item 4, very few children—0.1% (n = 179)—were identified as needing or receiving special therapy. For item 5, 1.9% (n = 4,324) were identified as needing or receiving treatment or counselling.

As expected, a higher percentage of boys than girls (19.3% versus 15.6%; p < .0001) were classified as CSHCN (Table 2). This was the case for each item except 4 (special therapy), for which no difference emerged, possibly because of the small sample size (n < 100) in each gender group. A non-linear relationship was observed for age, with the highest percentage of children classified as CSHCN at age 8. As children got older, medicine use, functional limitations, and the use of counselling services increased (p < 0.001). However, medical care use (item 2) decreased at older ages (p < 0.001), and the sample size for needing or receiving special therapy (item 4) was too small (n < 50) for age comparisons.

The prevalence of overnight hospital admissions was higher among the CSHCN group (5.6%), compared with the non-CSHCN group (0.7%; p < .0001) (Table 3). This difference was statistically significant for each Screener item. In fact, 70% of children who were hospitalized overnight were identified as CSHCN.

Children in the CSHCN group were more likely than those in the non-CSHCN group to be in a family that received premium subsidies: 35.3% versus 21.2% (p < .0001) (Table 4). This pattern held for each CSHCN Screener item.

A higher percentage of children in the CSHCN group than in the non-CSHCN group lived in the lowest income quintile neighbourhoods; however, the difference was small: 21.2% versus 19.7% (p < .0001)(Table 5). This was also the case for Screener items 2 (medical care use) and 3 (functional limitations).

Discussion

This study applied the CSHCN Screener tool—a non-categorical measure typically used with survey data—to administrative information from Population Data BC in order to identify children with health problems, as reflected in health service use. The aim was to determine what can and cannot be captured in administrative health data. To support the results, demographic and socio-economic characteristics of children classified as CSHCN and non-CSHCN were compared. Consistent with research based on survey data,Note 13,Note 22,Note 23,Note 26 application of the CSHCN Screener to administrative data revealed a higher prevalence of health care consequences among boys and among children from economically disadvantaged families (receiving premium subsidies and living in the lowest income quintile neighbourhoods). Children classified as CSHCN were also more likely to have been hospitalized.

According to this application of the Screener tool, 17.5% of children in the administrative data were identified as CSHCN. However, this study was limited to 6- to 10-year-olds in British Columbia. Health service use may be different for other ages and would likely differ by province. As well, the survey- or interview-based Screener covers both health care needs and health service use, but administrative data, by nature, capture only health service use. Nonetheless, the overall estimate is in line with American and Canadian surveys that identified 15% to 20% of children aged 0 to 17 as CSHCN.Note 13,Note 19,Note 26

For the medicine use item (1), this approach yielded 3.3% of children, which is considerably below other estimates. For example, using Canadian Health Measures Survey (CHMS) data for 2007 to 2009, Kuhle et al.Note 27 reported that 11% of 6- to 11-year-olds in British Columbia had taken medications (prescription and non-prescription drugs and natural health products) in the last month. More recently, based on 2007 to 2009 and 2009 to 2011 data from the same survey, Rotermann et al.Note 28 found that 12% of 6- to 14-year-olds had taken prescription medication in the last two days. Different definitions of item 1, such as counting days with two prescriptions filled as two days rather than one, and examining prescription drug use within a single 3-level ATC group, did not change the percentage of children identified in the administrative data by more than one percentage point. The lower prevalence in the present analysis may be because the study population was younger (ages 6 to 10) than those on which other estimates were based (ages 6 to 11 and 6 to 14), and because drug use was restricted to prescription medicines. Perhaps more important, the measure used here required continuity of prescription medicine use for at least nine months during a twelve-month period, as opposed to any medicine use, including prescription and non-prescription drugs and natural health products, in the last month or last two days.

For special therapy (item 4) and counselling (item 5), the percentages were lower than reported by survey-based studies.Note 13,Note 14,Note 19,Note 22 This may be due to lack of access to services and/or differences in data sources (medical claims versus population surveys). In particular, special therapy and counselling services are not covered by MSP, and so are under-reported in administrative data. In addition, MSP data pertain to services provided by physicians such as psychiatrists, but services provided by community and school-based professionals (for instance, speech therapists, specialized educators, and counselors) are not captured. And although the prevalence was comparable to those in other studies,Note 13,Note 21,Note 22 administrative data do not fully capture functional limitations and social participation.

Results for overnight hospital admissions, demographic, and community characteristics for each CSHCN Screener item were generally in the expected direction. For example, for most items, the percentage of boys was higher than the percentage of girls; the exception was special therapy (item 4), which may be due to small sample size. The results for age were mixed. The percentages for medicine use (item 1), functional limitations (item 3), and counselling (item 5) were higher for older than for younger children, but not for more medical care use (item 2) and special therapy (item 4). While this suggests that not all consequences increase with age, it may be that older children receive services not covered by the MSP, and so are not reflected in the administrative data. These findings warrant further research on factors that may influence children’s access to health services.

As expected, and consistent with the literature,Note 22,Note 29 the prevalence of overnight hospital admissions was higher for the CSHCN group and for each item. Also consistent with previous findings,Note 13,Note 30 a higher percentage of the CSHCN group were in families that received a premium subsidy; this held for each item.

This pattern was not replicated with the neighbourhood income variable. Although those in the CSHCN group who lived in neighbourhoods in the lowest income quintile had more medical care use and more functional limitations, the differences were not statistically significant for medicine use, special therapy, and counselling. The lack of neighbourhood differences may indicate equal medicine use or equal access, regardless of neighbourhood. Future research could evaluate potential biases (for example, those with limited access to health services) in these findings. For the special therapy and counselling items, the lack of neighbourhood differences may be due to the relatively small sample sizes (especially for special therapy: n < 50) or similar access to such services across neighbourhoods. Future research into children’s access to specialized health and mental health services could examine other neighbourhood characteristics, such as distance to services, hospitalsNote 31 and health care practitioners.

The findings of this analysis should be interpreted with some caution. For instance, the CSHCN Screener items used to assess functional limitations and the use of special therapies and counselling services may not be well-recorded in provincial administrative data. In addition, access to services, quality of life, and the severity of the health condition are not captured by the CSHCN Screener or in administrative data.

Conclusion

This study applied a non-categorical measure of child health, the CSHCN Screener, to province-wide administrative data. The availability of administrative data offers previously unexplored research opportunities in the area of child health. As a measure of service use, the CSHCN Screener may offer a non-categorical approach to identifying such children in administrative health data.

Acknowledgements

Funding for this study was provided by a Canadian Institutes of Health Research (CIHR) Operating Grant (MOP#102614) to Jamie C. Brehaut (co-PI) and Dafna E. Kohen (co-PI).

Date modified: