Health care quality measures for children and adolescents in Foster Care: feasibility testing in electronic records

Demands for information on the quality of pediatric preventive care have spurred investment in the development of quality measures designed to access the current state of children’s health care and ultimately define areas for improvement [1]. Specific quality measures for the foster care population, though, are largely untested [2, 3], even though youth entering foster care have greater physical, developmental, and mental health needs than their peers in the general population [4, 5]. Although children in foster care are known to have higher rates of social and medical morbidity, guidelines for the care of foster children are rarely adhered to in routine practice and may be difficult to measure [6‐10]. One of the most important ways to improve care and reduce poor long term health outcomes is through the development and testing of reliable quality measures for these high risk youth. To date, measurement has largely been dependent on labor-intensive chart reviews. Two other possible sources of quality data that might be extracted electronically exist for foster care children [11]. A clinical electronic health record (EHR) has potential to provide comprehensive and detailed patient-level data with electronic extraction on a large population. In addition, the Statewide Automated Child Welfare Information System (SACWIS) is a database used by protective services agencies to hold the official case records of children in care. Many states hold health data in their SACWIS records.

In order for these datasets to be useful, each of them would have to contain sufficient detail on health services to compare them against recommended guidelines, entry and exit dates for foster care to calculate eligibility, and demographic data for stratification. The primary objectives of this study were to identify quality measures, such as the appropriate number of well care visits, vaccinations and developmental screening for young children (ages 0–3) and adolescents (ages 12–18) in foster care and test whether the data required to calculate these measures can be feasibly interpreted within an EHR or within SACWIS. These two age groups were chosen because of the diversity of well care measures available for testing in both groups.

This work was performed as part of the Children’s Health Insurance Program Reauthorization Act (CHIPRA) Pediatric Quality Measures Program (PQMP), specifically as part of the National Collaborative for Innovation in Quality Measurement (NCINQ). Our approach considered three time periods of child welfare engagement: (1) entry into foster care, (2) ongoing foster care, and (3) foster home change or exit. [12‐24] We gathered input from a national advisory panel representing foster care alumni, national policy makers, state child welfare and Medicaid officials, health plan staff, and academic researchers.

Retrospective chart reviews were performed at Nationwide Children’s Hospital (NCH). For the study of children aged 0 to 3 years, we abstracted data from the time period of January 1, 2007- February 28, 2013 from children who met inclusion criteria: 1) In foster care (not including kinship care) within Franklin County, Ohio, and 2) at least one comprehensive well-care visit at a primary care physician (PCP) clinic or foster care specialty clinic at NCH. For the study of children aged 12–18 years, we abstracted data from the time period of January 1, 2009- October 31, 2013 with the same criteria. All extracted data was for care that occurred while the child was in foster care. We defined foster care as full-time care provided by an approved foster care family or group home and excluded any care provided by kin or close family friends (Table 1).

Chart reviews were performed by two staff members familiar with EHR data abstraction (Epic Systems Corporation, Wisconsin). An instruction document outlining the data elements and their common locations was created and used by both reviewers. Inter-rater reliability analyses were performed on the first 41 reviewed records of children within each age group to ensure reproducibility between data abstractors with excellent reproducibility [25, 26]. Study data were managed using REDCap data tool [27]. This study was approved by the NCH Institutional Review Board.

Because exact entry and exit dates for out of home care were often missing from EHRs, entry and exit dates were calculated in three different ways depending on the availability of data: 1) exact entry or exit dates were recorded whenever available, 2) the midpoint of an available date range (dates between health care visits wherein a child was documented to have entered/exited foster care) was used for entry or exit dates, and finally 3) the first well-care visit (or other documented healthcare encounter for adolescents) after entry was considered the entry date and the last well-care visit (or other documented healthcare encounter for adolescents) was considered the exit date when an exact date or date range was unavailable.

SACWIS is a “comprehensive automated case management tool that supports foster care and adoptions assistance case management practice.” [28] This system is intended to hold the official case record of all children currently or previously in out-of-home care in a state. Not all states use SACWIS, and there is substantial heterogeneity across states in the contents of SACWIS [28]. This study used data available in the Franklin County, Ohio SACWIS system in order to reflect the same population studied in the EHR review. A data extract from SACWIS was sent to the investigators containing the records of all children who were in custody during the study period and met inclusion criteria. Inclusion criteria were identical to those used for the EHR review, with the one exception that documentation of a well-care visit was not required. To determine whether data from the EHR and SACWIS could be reliably combined for analyses, we performed matching, using social security numbers when these were available in both the EHR and in SACWIS. As social security numbers were unavailable in one or both databases for approximately 70% of patients, when this number was unavailable we also considered records to be from the same child if they matched on all four of the following criteria: last name, first name, date of birth, and gender. A last name in the EHR was considered to match a last name in SACWIS if the first four characters were identical. Both the primary name and alias listed in SACWIS were considered. A first name in the EHR was considered to match a first name in SACWIS if the first four characters, either with or without symbols, were identical. Again, both the primary name and alias listed in SACWIS were considered.

Continuous variables were described with medians and interquartile ranges (IQR) as none were normally distributed. Categorical variables were described using frequencies and percentages. Quality measures were calculated in either the entire young child or the entire adolescent study cohort. We chose to review a sample of 400 EHRs for both the young child and adolescent EHR studies. The medical record numbers of all included children were sorted randomly such that the children whose charts were reviewed were a random sample of all children who could have been included.

In order to provide the most flexible information on the feasibility of obtaining the proposed quality measures from our data sources, two methods were used to calculate the proposed quality measures: proportions and observed-to-expected ratios. Most quality measures of pediatric healthcare focus on the former. Unfortunately, calculations of such often require steady denominators with fixed lengths of follow-up such as the number of children screened over the number of children eligible who were tracked for a full year. Because foster children cycle in and out of care, the denominator calculations may be ineffective in describing this unstable population.

Several quality measures were calculated as observed (received) to expected (recommended) ratios (O/E ratios) to better describe the actual quantity of recommended health care that was received by individual children, whereas the proportion measures indicate the percentage of children in the study cohort that received all recommended care. Weighted mean O/E ratios were calculated wherein each child’s individual O/E ratio was weighted by his or her total time spent in foster care during the study period. This weighting was performed because it enabled children who spent more time in foster care to contribute more to the calculated O/E measures. We used the AAP Recommendations for Preventive Pediatric Health Care as the guideline for determining the expected, or recommended, number of well-care visits and developmental screenings in the young children and the recommended number of well-care visits, BMI measurements, drug use assessments, alcohol use assessments, sexually transmitted infection screenings, and depression screenings in the adolescents [24]. For each measure, an individual O/E ratio was calculated for each child and then a weighted average of these individual O/E ratios was calculated to provide an appropriate average O/E for the entire cohort.

Continuous variables were described with medians and interquartile ranges (IQR). Categorical variables were described using frequencies and percentages. We also attempted to extract data on the same types of health care encounters that were examined in the chart reviews. SAS version 9.3 (SAS Institute Inc., NC) was used to analyze all data.

Documentation of data important to the tracking and optimization of the health care of children and adolescents in foster care is frequently incomplete and difficult to find in either EHRs or SACWIS in our patient population. However, despite their limitations, EHRs and SACWIS can be useful data sources for the calculation of some important measures of quality of care in the foster care population, and would be even more useful if certain important data elements were more consistently available and easily extractable from each database. Alternatively, individual level matching across platforms may allow for the optimal methods by which to assess these measures.

Manual review to calculate our proposed foster care quality measures was laborious. The majority of information was located in free text fields and scanned documents rather than structured fields. Many important data elements, specifically the reasons for initial entry into foster case and the entry and exit dates from foster care, were often missing, and this lack of documentation proved to be limiting factors in data abstraction. The accuracy of nearly all of our proposed health care quality measures is contingent upon this critical information. Similar issues were identified for SACWIS data. While demographics and entry and exit characteristics were found in discrete fields, all other information of interest to this study was located in free text fields. This required study investigators to visually examine text notes. Even after this task was performed, it was found that the quantity of information and detail on medical care in SACWIS was far less than from the EHRs.

Identifying the best method to calculate the proposed quality measures revealed the complexity of EHR data abstraction and quality measure development for children in foster care. For example, the Healthcare Effectiveness Data and Information Set (HEDIS) Well-Child Visit measure at 15 months of age would have minimal utility in the foster care population as it requires continuous enrollment for 12 months prior to age 15 months as an inclusion criteria [29]. The O/E ratios examined in this study seemed to be a viable option for the calculation of quality of care measures in the foster care population, primarily because every child can be included in the calculation of these measures regardless of their length of stay or number of episodes in foster care. In addition, children who spend more time in foster care appropriately contribute more to these measures than children who spend less time in foster care. The results in Tables 3 and 5 indicate that select quality measures appear better when calculated using O/E ratios rather than proportions, namely because all health care events contribute to the O/E measures whereas with the proportion measures, a child is counted in the numerator only if the ideal number of events of interest occurred. Admittedly however, the greater mathematical complexity of the weighted average O/E measures, compared to simple proportions, may limit their widespread use.

Considering the challenges we encountered in this study, modification of current EHRs, the use of another data source, or combination of data sources may improve the feasibility of foster care quality measures. An ideal EHR format specific to children was recently proposed [30]. The format provides specific elements and requirements that could be added to current EHRs to enhance the care of children, especially those enrolled in Medicaid and in the care of child welfare [30]. These recommendations include system capacity to store and display 1) whether the child has ever been in out-of home care 2) information about the dates of the out-of-home care and 3) information on the child’s history of abuse or neglect. In addition, the SACWIS data system used by child welfare agencies could also be useful. Although its current use varies by state, it is intended to be a comprehensive database that supports the efforts of case workers to assist children in out of home care [28]. The availability of exact dates of entry into foster care and exit from foster care in SACWIS and the availability of accurate data on health care received in the EHR could, together in a combined database, enable the calculation of more accurate health care quality measures than those presented here. However, a higher match rate than we found would be necessary to make such a combined database useful.

Extraction of data to test foster care quality measures is not currently feasible in a single institution EHR, even though we conducted this study at a large, free-standing children’s hospital with a longstanding commitment to electronic health records, nor is it feasible in a metropolitan county’s SACWIS data. Most proposed quality measures tested did not achieve high adherence as recommended by current guidelines, but it is difficult to tell to what extent missing data elements such as entry and exit dates contributed to these results. Because the quality of information is important to improve patient care, testing foster care quality measures in SACWIS or an augmented EHR that utilizes the children’s EHR format may be a better alternative, and subsequently may yield more reliable results [31].