|Home | About | Journals | Submit | Contact Us | Français|
Primary care pediatricians increasingly care for children’s mental health problems, but little is known about practice-level variation in diagnosis and psychotropic medication prescribing practices.
This retrospective review of electronic heath records from 43 US primary care practices included children aged 4 to 18 years with ≥1 office visit from January 1, 2009, to June 30, 2014. We examined variability in diagnosis and psychotropic prescribing across practices using logistic regression with practice fixed effects and evaluated associations of the availability of colocated or community-based mental health providers or the proportion of children in foster care with diagnosis and prescribing using generalized linear mixed models.
Among 294748 children, 40932 (15%) received a mental health diagnosis and 39695 (14%) were prescribed psychotropic medication. Attention deficit/hyperactivity disorder was most commonly diagnosed (1%–16% per practice). The proportion of children receiving any psychotropic medication (4%-26%) and the proportion receiving ≥2 medication classes (1%-12%) varied across practices. Prescribing of specific medication classes also varied (stimulants, 3%–18%; antidepressants, 1%–12%; α-agonists, 0%–8%; second-generation antipsychotics, 0%–5%). Variability was partially explained by community availability of psychiatrists (significantly higher odds of a diagnosis or prescription when not available) but not by colocation of mental health professionals or percentage of children in foster care.
The prevalence of mental health diagnosis and psychotropic medication prescribing varies substantially across practices and is only partially explained by psychiatrist availability. Research is needed to better define the causes of variable practice-level diagnosis and prescribing and implications for child mental health outcomes.
Primary care pediatricians’ comfort in caring for children with mental health conditions varies widely. Although previous studies have documented variability in diagnosis and prescribing by geographic region and provider type, primary care practice–level variability has not been examined in depth.
Through a detailed review of electronic health record data, we found substantial practice variation in mental health diagnosis and psychotropic medication prescribing. Community availability of psychiatrists, but not colocation of mental health providers, partly explained this variation.
In 1991, the American Academy of Pediatrics (AAP) policy statement, “The Pediatrician and the ‘New Morbidity,’” highlighted pediatricians’ expanding role in managing children’s psychosocial problems.1 In the years since adoption of the statement, primary care pediatricians have cared for an increasing number of children with mental health conditions.2 Currently, a majority of psychotropic medications are prescribed by primary care providers rather than specialists.3,4 Surveys have found that many pediatricians felt it was their responsibility to identify children with mental health concerns but not to manage these conditions,5–7 with the exception of attention deficit/hyperactivity disorder (ADHD).5,8 Lack of confidence,6,7,9 knowledge/skills,7,9,10 or time7,9,10 are barriers that limit pediatricians’ ability to care for children with mental health issues.
In an effort to help primary care pediatricians manage behavioral, developmental, and other mental health conditions, the AAP released toolkits providing resources for managing ADHD,11 autism,12 and other mental health concerns.13 However, these kits are not used consistently,14 and mental health care among children and adolescents continues to vary widely in the United States. Prior studies have identified more than twofold variation in the rate of mental health service use across states.15–17 Variation is even greater among children in the child welfare system, particularly foster care.18,19 Whereas patient characteristics have been linked to variation in mental health care,19,20 several studies have demonstrated variation in mental health care that was explained by geography rather than patient characteristics.15,21 These results are especially concerning given known regional variation in unmet mental health care needs.22
Although prior studies have documented variability by geographic setting, they have not focused on primary care practice–level variability, an area of importance since practices are the microsystems through which care is delivered.23 Several practice-level characteristics might affect the primary care delivery of mental health services. In response to family difficulties in accessing mental health specialists,24,25 primary care practices increasingly include colocated mental health providers,26–30 a model that may improve access, identification, and treatment.26,31 Second, practices with higher numbers of patients in foster care may manage more mental health care owing to prevalent psychotropic medication use.18,32 Finally, practices located in communities where mental health providers, particularly psychiatrists, are not available may need to fill this gap by providing psychotropic medication management.33 These factors may drive practice differences; however, their association with practice-level variability has not been examined.
To address this gap, we investigated practice-level variability in mental health diagnosis and psychotropic medication prescribing for children cared for by primary care pediatricians in 2 practice-based research networks. We examined whether variability in prescribing and diagnosis was explained by practice-level characteristics including colocated or available community mental health providers (including psychiatrists) and the proportion of children in foster care. We hypothesized that diagnosis and prescribing would be more common at practices with colocated mental health providers, a higher proportion of children in foster care, and in communities without psychiatrists. We also hypothesized that, given the growing comfort with ADHD treatment among pediatricians,34 practice variation would be smaller for the diagnosis and management of ADHD versus other conditions.
This study was conducted within 2 electronic health record (EHR)-based practice-based research networks: the Pediatric Research Consortium (PeRC) of the Children’s Hospital of Philadelphia, a hospital-owned network including 24 practices (29 sites) serving >200000 children,35 and the Electronic Pediatric Research in Office Settings (ePROS) network of the AAP, including 19 practices (21 sites) using 7 different EHR vendors and serving ~90000 children in 15 states.35 All practices in the PeRC network participated, and ePROS practices joined the network because of an interest in participating in EHR-based research; practices were not recruited for this specific study. We conducted a retrospective review of EHRs from these 43 practices to identify a cohort of children aged 4 to 18 years with visits from January 1, 2009, to June 30, 2014. As the child was the unit of analysis for this study, we created a child-level dataset that included all children meeting the inclusion criteria.
The primary endpoints included mental health diagnosis and psychotropic medication prescribing. Mental health diagnoses were identified with codes from the International Classification of Diseases, Ninth Revision, and children were considered to have a diagnosis if there was evidence of that diagnosis in the EHR at any time during the study period.
The following diagnostic categories were included: ADHD (314.0–314.9), autism spectrum disorder (299.00–299.99), schizophrenia (295.00–295.99), bipolar (296.00–296.10, 296.36–296.89), anxiety (300.00–300.29, 301.4), conduct disorder (312.00–312.89), oppositional defiant disorder (313.81), and depression (311, 296.20–296.35). Children who received a diagnosis of seizure disorder (345.00–345.91) were excluded to prevent confusion between treatment with antiepileptics and mood stabilizers.
Psychotropic medication prescribing was defined as evidence of any psychotropic medication prescription from EHR orders or medication lists for each child at any time during the study period. Medications were entered into the EHR when prescribed or when abstracted into the record by clinicians at participating practices. The prescription of psychotropic medications was identified from the National Drug Code or Generic Product Identifier specific to each drug, as well as by examining medication names. We classified psychotropic medications into 7 therapeutic classes as follows: stimulants, selective norepinephrine reuptake inhibitors (atomoxetine), α-agonists (clonidine and guanfacine), antidepressants (selective serotonin reuptake inhibitors, serotonin-norepinephrine reuptake inhibitors, tricyclics), second-generation antipsychotics (SGAs), anxiolytics (benzodiazepines, others), and mood stabilizers (carbamazepine, valproic acid, gabapentin, lamotrigine, oxcarbazepine anticonvulsants, and lithium).
Patient-level independent variables were extracted from the EHR. Child gender and age at the first visit during the study period (categorized as 4–11 years and 12–18 years) were available for all children. Race was inconsistently available (missing for 29%) and highly correlated with practice. Because it was difficult to separate the effects of race and the effects of practice, race was included only in sensitivity analyses. Insurance status was available only for PeRC patients (missing for 39% overall) and was not included in multivariable models.
Practice-level variables, assessed using survey items developed through review of the 2013 AAP 85th Periodic Survey34,36 and iterative consultation with pediatric primary care practitioners and behavioral health specialists for face validity, were collected from the lead physician at each practice (Supplemental Appendix). Clinicians reported whether the practice included colocated mental health practitioners (yes/no, separately for each of the following types of providers: child psychiatrists, child psychologists, developmental behavioral pediatricians, substance abuse counselors, mental health counselors, social workers, and adult psychologists/psychiatrists). The survey also assessed the perceived availability of mental health practitioners within the practice’s community (not available, somewhat available, and very available, separately for each of the above provider types) and the estimated proportion of patients who were in foster care. We also recorded the practice setting (urban, suburban, or rural) based on practice self-report. Finally, we calculated the number of patients cared for and the average age of patients at the practice.
The study cohort and practice characteristics were described using proportions. We first calculated the proportion of children with any, ≥2, and specific mental health diagnoses. Medication prescribing in the entire study population was examined by calculating the proportion of children receiving any, ≥2, and specific classes of psychotropic medication. Differences in diagnosis and prescribing by child age and gender were described.
We examined rates of diagnosis and prescribing separately within each primary care practice. We used multivariable logistic regression models with a fixed effect for practice to statistically assess variation in each diagnosis and medication class while controlling for patient age and gender. Sensitivity analyses considered race as a covariate, and results were consistent (data not shown). To assess whether variability was significant, likelihood ratio tests compared these models with reduced models that did not include the practice (P < .05 significant).
We then examined the association of practice characteristics with diagnosis and prescribing, including colocation of mental health professionals (overall, and among only those with prescribing authority [psychiatrists and developmental behavioral pediatricians]); perceived community availability of psychiatrists and other mental health providers (not at all available versus somewhat/very available); percentage of patients in foster care (≥6% vs 0%–5%); practice setting (urban/rural versus suburban); practice size; and age distribution of the practice. To examine practice characteristics, a distinct approach from that described above was required. Generalized linear mixed models with a logit link, practice characteristics as fixed effects, and practice site as a random intercept (meqrlogit in Stata) examined whether practice characteristics explained variation in diagnosis and prescribing, adjusting for patient-level covariates. Interactions were tested between practice characteristics and child age and gender. When interaction terms were significant, stratified odds ratios were calculated. However, because patterns were consistent between strata even when interaction terms reached statistical significance, only overall results are presented.
All analyses were conducted in Stata version 13.1 (StataCorp, College Station, TX) and SAS version 9.3 (SAS Institute, Cary, NC). The institutional review board at the AAP approved this study, and the institutional review board at the Children’s Hospital of Philadelphia determined that this study was not human subjects research.
The study cohort included 294748 children from 43 primary care practices. During the study period, 40932 children (15%) had ≥1 mental health diagnosis, and 8719 (3%) had >1 diagnosis (Table 1). Consistent with national data,16 the most common diagnosis was ADHD (25307 children, 9% overall and 62% of those with a mental health diagnosis). In total, 39695 children (14%) received a prescription for ≥1 psychotropic medication during the study period. The most commonly prescribed medications were stimulants (9%) and antidepressants (4%). Consistent with previous findings,16 boys were more likely to receive any psychotropic medication than girls (17% compared with 10%), as were those aged 12 to 18 compared with those aged 4 to 11 (17% compared with 12%) (P < .001 for both). Overall, 14397 children received prescriptions from ≥2 medication classes (5% of the entire cohort and 36% of those on any psychotropic medication), most commonly stimulants and α-agonists (5123, 36% of those on ≥2 medication classes).
Sixteen practices (37%) reported onsite mental health professionals, most commonly child psychologists (9 practices, 21%) and social workers (8 practices, 19%) (Table 2). Clinicians reported low levels of availability of mental health professionals in their communities, particularly for child psychiatrists and developmental behavioral pediatricians [12 (28%) and 13 (30%) practices reported these clinicians were not at all available, respectively]. Additionally, only 8 practices (19%) reported that foster care patients made up ≥6% of their total patient population. The number of children per practice included in the analysis ranged from 575 to 21549 (median 5662), and the average age in practices ranged from 8.2 to 10.9 years (median 10.0). One practice did not complete the survey at all, and 2 others did not respond to specific sections as noted in Table 2.
There was significant variability in all mental health diagnoses and psychotropic medication classes across practices (P < .001 for all comparisons based on likelihood ratio tests) (Table 3). The proportion of children with any mental health diagnosis ranged from 2.3% to 22.2%. Contrary to our hypothesis, we observed substantial variability in the proportion of children receiving an ADHD diagnosis (1.3%–15.8%) across practices relative to other diagnoses. Differences in other diagnoses were smaller, but still meaningful: 0.5% to 7.7% for anxiety, 0.0% to 4.8% for depression, 0.2% to 3.0% for autism, 0.0% to 3.1% for conduct disorder, and 0.0% to 2.3% for oppositional defiant disorder. Bipolar disorder was uncommon (≤1.0%).
As with diagnosis, the proportion of patients receiving any psychotropic medication varied substantially across practices (Fig 1, Table 3), ranging from 4.3% to 25.8%. In addition, there was >20-fold variation in the proportion of patients receiving prescriptions from ≥2 medication groups, with a range of 0.5% to 11.5%. Practice variability in prescribing was observed for the 4 most commonly prescribed medication classes in this sample: stimulants, antidepressants, α-agonists, and SGAs. Stimulant prescribing was the most variable (range 3.3%–17.9% compared with 0.9%–11.8% for antidepressants, 0.1%–8.0% for α-agonists, and 0.1%–5.0% for SGAs). Atomoxetine, anxiolytics, and mood stabilizers were uncommon across sites (maximum of 3.5%).
Variability in diagnosis and prescribing was partially explained by perceived community availability of psychiatrists. Children at practices reporting that psychiatrists were “not at all available in the community” had significantly higher odds of any mental health diagnosis, ≥2 diagnoses, ADHD, and depression compared with children at practices reporting that psychiatrists were available (Table 4). Children at practices where clinicians reported no available community psychiatrists also had significantly higher odds of psychotropic medication prescribing. Results were statistically significant for any psychotropic medication, ≥2 medication classes, stimulants, antidepressants, and α-agonists. However, perceived community availability of child psychologists was not associated with diagnosis and prescribing.
Although significant, differences in perceived community availability of psychiatrists did not entirely explain the variability in diagnosis and prescribing observed. Likelihood ratio tests indicated that practice variability was still statistically significant after adjustment for community availability of psychiatrists for all models (P < .001).
Contrary to our hypothesis, colocation of mental health providers was not associated with diagnosis or prescribing (Table 4). Results were similar when we included only colocated professionals with prescribing authority. Similarly, a high prevalence of foster care children in the practice was not associated with diagnosis or prescribing (except for anxiety diagnosis), nor were practice setting, practice size, or age distribution of patients at the practice.
Previous research has focused on the diagnosis and treatment of mental health problems at the population level as well as individual pediatrician willingness to treat mental health concerns. Using a database of 43 practices’ EHRs covering a 5-year period, we identified significant variation in mental health care at the practice level. We found the highest levels of variability for ADHD diagnosis and prescribing of psychotropic medications, especially stimulants. This variability was partially explained by the perceived availability of psychiatrists in the communities in which practices were located. Children at practices reporting no availability of psychiatrists were more likely to have a diagnosis or prescription recorded in their primary care EHR, a pattern that was not observed for other types of mental health professionals. Other practice characteristics, including the colocation of onsite mental health professionals and the prevalence of foster care children in the practice, were not associated with diagnosis nor prescribing.
The high level of variability observed in this study at the practice level is consistent with prior research assessing patterns in the population. A national survey found more than twofold variation in the use of mental health care across states, from 5% to 11%,15 whereas a study in Washington State found that the percentage of Medicaid-eligible children receiving mental health care ranged from 3% to 8% throughout the state.21 We found even greater variation in psychotropic medication prescribing in our cohort of children in primary care, although not as large as several national studies of children in the child welfare system.18,37 In our study, this significant variability was only partially explained by community availability of psychiatrists, and was not associated with other characteristics we hypothesized would increase diagnosis and prescribing.
Shortages in child and adolescent psychiatrists, particularly common in rural areas or in communities with more children living in poverty,38 may make it necessary for pediatricians in certain areas to treat mental health conditions beyond ADHD, a trend that has been described previously.39 In a survey of nearly 3000 primary care providers, two-thirds reported they were unable to obtain mental health services for at least some of their patients, and in counties with few psychiatrists available, primary care providers were more likely to report shortages of providers as a barrier to accessing services.40 Additionally, previous research has documented that the availability of nonpharmacological treatments such as behavior therapy is highly variable throughout the country (33%–61% among children with current ADHD as of 2009–2010)41 and depends on system-level factors such as insurance coverage and state policies. In this context, our study suggests that pediatricians may, out of necessity, prescribe psychotropic medications to manage a larger percentage of mental health care in areas lacking available psychiatrists. This association was not observed for psychologists. Nonetheless, psychologists, who may be more focused on prevention and early intervention and less likely to formally diagnose children, may impact child well-being without influencing diagnosis and prescribing.
Colocation of mental health providers in primary care practices has been proposed to remove barriers to access.26,28,30,42,43 Previous studies have found that colocation improved access to mental health care30,44 and engagement with mental health care providers31,45 and improved patient satisfaction and outcomes.46,47 We hypothesized that colocation would result in increased diagnosis and prescribing by bringing expertise with these conditions to the practice and potentially drawing additional children needing these services. However, our finding that colocation was not associated with diagnosis or prescribing rates suggests that care by nonprescribing mental health professionals (psychologists and social workers) may have limited impact on practice-level prescribing. Additionally, even with colocation, barriers such as financial differences in reimbursement for medical and mental health services,48,49 difficulties with information sharing,49 differing expertise,48,49 and limited hours may impede integration.
The high level of practice variability in our study also suggests that patients with mental health conditions may have different care experiences by practice due to differences in clinician familiarity with treating children with behavioral health problems. This variability underscores the need to continue to evaluate and implement approaches to support practices in treating affected children. As community availability of psychiatry only partially explained the wide variation in diagnosis and prescribing, additional factors that may drive variability require future study. For example, primary care providers’ level of agreement with current guidelines, perceived self-efficacy in diagnosing or treating particular conditions, training, relationships with schools, and reimbursement from insurers might affect prescribing practices.50
This study had several limitations. Although primary care practices typically reconcile diagnoses given and medications prescribed by outside providers, medications and diagnoses may have been underestimated if physicians were unaware of or did not document them, children did not attend recommended preventive visits, or EHRs did not provide a shared record. However, given high levels of variability in ADHD diagnosis and stimulant prescribing that are primarily managed in primary care, it is unlikely that the observed variability is a documentation artifact. The data reflect psychotropic medication prescribing, but not necessarily use. Furthermore, as a majority of practices were in one region of the country, we were unable to test for regional variation. Also, information on prevalence of mental health providers was based on practice leader report but may not reflect actual availability. In addition, it is unknown whether any practices had systematic screening programs to identify behavioral or emotional disorders, which might have resulted in higher rates. Because we conducted a child-level analysis and children often received care from multiple providers, we were unable to examine variability between clinicians within practices. Finally, as this study focused on practice-level variability and practice characteristics, we were unable to focus on several important areas relevant to mental health care for children: trajectories of medication use or episodes of care among children over time (since we considered whether diagnoses or prescriptions were received at all by a child over the study period), child-level concurrent polypharmacy, use of community mental health or psychiatric consultation services, and child mental health outcomes. Such topics should be explored in the future.
Mental health diagnosis and prescribing for children in primary care are highly variable. Community availability of psychiatrists only partly explained this variation. Other factors, including presence of onsite mental health professionals or prevalence of foster care children, were not associated with variation in this study. Given the magnitude of the observed variability, further study is needed to better understand the causes of variable practice-level diagnosis and prescribing and the implication for child mental health outcomes.
We thank the practitioners, patients, and their families from PeRC and ePROS.
The ePROS pediatric practices that participated in this study are listed by AAP Chapter. California: Shasta Community Health Center (Redding), practice of Mark M. Simonian, MD (Clovis); Colorado: Fort Collins Youth Clinic PC (Fort Collins); Georgia: The Pediatric Center (Stone Mountain), Roswell Pediatrics (Cumming); Indiana: Jeffersonville Pediatrics (Jeffersonville); Kentucky: Union Pediatrics (Union); Maryland: Main Street Pediatrics (Towson); New Jersey: Delaware Valley Pediatric Associates PA (Lawrenceville), PASE Healthcare PC (Millburn); New York: East End Pediatrics (East Hampton); Missouri: Priority Care Pediatrics LLC (Kansas City); Ohio: Oxford Pediatrics and Adolescents (Oxford); Oklahoma: OUCP Sooner Pediatric Clinic (Oklahoma City); Oregon: Childhood Health Associates of Salem (Salem); Tennessee: Plateau Pediatrics (Crossville), Cool Springs Pediatrics PLLC (Franklin); and South Carolina: AnMed Health Children’s Health Center (Anderson).
Ms Mayne drafted the initial manuscript; Ms Mayne and Drs Ross and Localio contributed to the analysis and interpretation of the data; Dr Ross critically reviewed the manuscript; Mr Song, Ms McCarn, Ms Steffes, Ms Liu, and Drs Margolis, Azuine, Gotlieb, Grundmeier, Wasserman, and Fiks contributed to the acquisition of the data; Mr Song and Drs Grundmeier, Wasserman, and Fiks contributed to the analysis of the data; Mr Song, Ms McCarn, Ms Steffes, Ms Liu, and Drs Margolis, Azuine, Gotlieb, Grundmeier, Leslie, Localio, Wasserman, and Fiks critically reviewed the manuscript; Drs Grundmeier, Wasserman, and Fiks contributed to the conception and design of the study; Drs Grundmeier, Leslie, Wasserman, and Fiks contributed to the interpretation of the data; and all authors approved the final manuscript as submitted.
This information or content and conclusions are those of the authors and should not be construed as the official position or policy of, nor should any endorsements be inferred by, the Health Resources and Services Administration, the US Department of Health and Human Services, or the US government.
FINANCIAL DISCLOSURE: Dr Fiks is the co-principal investigator of an independent research grant from Pfizer for work in ADHD; the other authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Supported by the Health Resources and Services Administration (HRSA) of the US Department of Health and Human Services (HHS) under grants R40MC24943 (“Primary Care Drug Therapeutics CER in a Pediatric EHR Network"), UB5MC20286 (“Pediatric Primary Care EHR Network for CER"), and UA6MC15585 (“National Research Network to Improve Child Health Care"). Funding was provided, in part, by a grant from the Eunice Kennedy Shriver National Institute of Child Health and Human Development under the Best Pharmaceuticals for Children Act. Funded by the National Institutes of Health (NIH).
POTENTIAL CONFLICT OF INTEREST: Dr Fiks is the co-principal investigator of an independent research grant from Pfizer for work in ADHD; the other authors have indicated they have no potential conflicts of interest to disclose.