Patterns of Antipsychotic Prescribing by Physicians to Young Children

We obtained monthly physician-level prescribing information from IMS Health’s Xponent database, which directly captures over 70% of all U.S. prescriptions filled in retail pharmacies and utilizes a patented projection methodology to represent 100% of prescriptions filled in these outlets. IMS Health provided Xponent data on all U.S. physicians classified as psychiatrists in the American Medical Association Masterfile (N=29,857) and a random sample of 5% of physicians classified as family medicine physicians (N=1,856) who wrote at least ten antipsychotic prescriptions per year from 2008 to 2011. The Xponent data include information on the payer (Medicaid fee-for-service [FFS], Medicare, commercial, or cash) and the age of the patient by IMS-provided category (0–9, 10–19, 20–64, ≥65). For the purposes of this study, we defined young children as those between the ages of zero and nine. Adults were defined as persons between the ages of 20 and 64. Because the data are obtained from pharmacy transactions and not from medical claims, they lack patient clinical information, for example, diagnosis codes. We linked the Masterfile data on physician characteristics (age, sex, geographic location, training institution, and practice setting) to the Xponent prescribing data by using a prescriber ID. As a result, these data included detailed information on physician characteristics and comprehensive physician-level data on prescribing patterns across all payers for over 31,000 psychiatrists and family medicine physicians who prescribed antipsychotics.

To identify physician characteristics associated with prescribing antipsychotics to young children, we estimated a logistic regression model examining whether having prescribed an antipsychotic to a young child at least once during 2009–2011 (using 2008 data to measure baseline variables) was a function of physician sex; age (≤39, 40–49, 50–59, or ≥60); practice setting (solo or two-person; group practice; other, such as U.S. Department of Veterans Affairs, military hospitals, or U.S. Public Health Service; medical school; health maintenance organization; or no classification available); any hospital practice; rural practice; U.S. Census region (New England, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain, Pacific, or Middle Atlantic); graduation from a top-25 medical school, as ranked by U.S. News and World Report in 2011; foreign medical graduate; quartile of total 2008 antipsychotic prescription volume (the total number of antipsychotic prescriptions for adults or children); and share of total antipsychotic prescriptions in 2008 paid by a Medicaid FFS program, by Medicare or a commercial payer (including Medicaid managed care plans), or by cash. Because almost 90% of child and adolescent psychiatrists had prescribed an antipsychotic for a young child at least once, we did not include specialty indicators.

Next, we compared prescribing patterns for young children versus adults among the subset of physicians who had at least one prescription for an antipsychotic for both age groups (N=13,214). We were interested in whether prescribers chose a different, and possibly narrower, set of options when prescribing an antipsychotic to children, given the limited number of products with FDA approval for use by young children. We first compared the share of all antipsychotic prescriptions represented by each medication for patients in the two age groups. We then examined the number of antipsychotic medication ingredients prescribed for patients in the two groups by all physicians, by specialty, and by quartile of total antipsychotic prescribing volume in 2008. Because of the very low use of first-generation antipsychotics among young children, all first-generation antipsychotics were treated as a single ingredient. Each second-generation agent was treated as a separate ingredient. [A list of first-generation antipsychotics included in the analyses is available as an online supplement to this article.]

We also identified factors associated with prescribing of antipsychotics for which there were no FDA-approved indications for young children. As noted above, only two second-generation antipsychotics, risperidone and aripiprazole, had an approved indication for young children [see online supplement]. Four first-generation drugs—chlorpromazine, haloperidol, prochlorperazine, and trifluoperazine—had approved indications for young children, and prescriptions for these were categorized accordingly, although these drugs were rarely used among young children.

We estimated a hierarchical binomial regression model of the number of prescriptions for antipsychotics with no approved indications for young children that physicians had written for children ages zero to nine. We assumed that the number of prescriptions for antipsychotics with no approved indications for young children written by each physician for a child in that age group arose from a binomial distribution characterized by a physician-specific probability (p). We modeled the log odds of the probability of these prescriptions as a function of the same variables that were used in the logistic regression model described above, with two additions for specialty (child or adolescent psychiatry, other psychiatry, or family medicine) and a physician-specific random effect. The random effect was assumed to be normally distributed with a mean of zero and an unknown variance of τ²; the random effect accounted for clustering of prescriptions written by each physician as well as different physician prescribing volumes. We computed odds ratios (ORs) and corresponding 95% confidence intervals (CIs) to indicate the association between off-label prescribing of antipsychotics for young children and physician characteristics. To characterize the degree of between-physician variation in the probability of writing these prescriptions after adjustment for physician characteristics, we determined the odds among young children of receiving such a prescription from a “high” prescriber versus a “low” prescriber. A high prescriber was operationalized as a prescriber whose prescribing was one standard deviation (τ) above the mean of the physician random effects. To determine whether physicians who wrote few prescriptions affected our findings, we restricted the sample to physicians with at least 30 prescriptions for an antipsychotic and repeated the binomial regression analysis.

This study was approved by the institutional review boards of Harvard Medical School and the University of Pittsburgh. All analyses used SAS, version 9.4.

Of the 31,713 physicians, 13,374 (42.2%) wrote at least one antipsychotic prescription to a patient between ages zero and nine during the three-year period. These physicians differed along a number of dimensions compared with prescribers with no prescriptions for this population (Table 1).

Using all-payer, physician-level prescription data on psychiatrists and family medicine physicians who prescribed antipsychotic medications over the period 2009 to 2011, we found that over 40% of prescribers had written at least one antipsychotic prescription for a child between the ages of zero and nine. Prescribing antipsychotics to young children was more common among younger physicians, those practicing in rural areas, those with higher total antipsychotic prescribing volume, and those with a higher share of prescriptions paid by Medicaid FFS programs. It was less common among physicians in settings outside group practices and in two-person or solo practices, physicians who graduated from a top-25 medical school, and physicians with a high share of antipsychotic prescriptions paid by cash. Although three-quarters of prescriptions for young children were for the two drugs with an FDA-approved indication for children in that age range, almost two-thirds of physicians who prescribed an antipsychotic to young children had written at least one prescription for a medication with no FDA-approved indication for children between the ages of zero and nine. That was in spite of the fact that physicians tended to prescribe from a narrow set of antipsychotics (typically two) for young children.

Although previous studies using patient-level claims and survey data have documented rates of antipsychotic use among young children (3–7), little is known about the characteristics of physicians who prescribe these medications to young children or about the patterns of prescribing in this age group. Our results show that physicians who treated both adults and children appeared to tailor their prescribing to some extent on the basis of age group, adopting what may be a more conservative approach for young children. Compared with prescribing practices for adults, when treating young children physicians were more likely to prescribe an antipsychotic with an FDA-approved indication for that age group. They were also more likely to use a much smaller number of ingredients for young patients compared with adult patients.

A number of strategies, including prior authorization and step therapy requirements (39–44), tiered formularies (45,46), monitoring and feedback (such as computerized alerts, reminders, audit and feedback) (47), academic detailing (48,49), and policies restricting detailing efforts by pharmaceutical manufacturers (50), have proven effective at influencing physician prescribing behavior in general. Payers and health plans could consider applying these approaches to antipsychotic prescribing for young children in an effort to remind prescribers of the risk-benefit tradeoffs associated with antipsychotic use among children. Some antipsychotic prescribing among young children may occur as a result of limited access to nonpharmacologic evidence-based treatments for challenging symptoms, particularly disruptive behaviors, as is often the case with severely hyperactive children with ADHD (26,36). This possibility points to the need for broader interventions focused on the competencies and geographic distribution of the mental health workforce and financing of mental health services (37,38,51,52). It is significant in this regard that federal and state agencies have already launched initiatives aimed at improving the appropriateness of child psychotropic prescribing, especially for children in foster care (30,53). New Healthcare Effectiveness Data and Information Set measures focused on the safe prescribing of antipsychotics for children have also drawn increased attention to this issue from payers, health plans, and provider organizations.

There were several limitations to our analysis. First, our data set lacked information on the specific indication for which a physician prescribed an antipsychotic as well as information on dosing, duration of medication therapy, and whether other therapies had been tried before an antipsychotic was prescribed. Previous studies of antipsychotic use among young children have relied on claims data (often from a single state Medicaid program or private insurer) (3,4,6,7,31) or the National Ambulatory Medical Care Survey (NAMCS), which asks a sample of physicians about patient visits over a one-week period (5,32,33). Although claims data provide information on diagnoses and service use, some claims data sets do not include physician identification numbers or information on physician characteristics. In addition, claims data from an insurer or Medicaid program would not be representative of the universe of a physician’s patients, given that most physicians treat patients with a variety of payers. Although NAMCS provides detailed information on a subset of visits, it cannot be used to follow physicians or patients over time.

The data used for this study allowed us to fully characterize a physician’s prescribing to young children over time and to identify physician characteristics associated with various prescribing patterns. Second, the Xponent data provide two age categories for children—ages zero to nine and ages ten to 19—so we were obliged to define young children as those between the ages of zero and nine. Third, we were unable to determine whether a given prescription was for an on- versus off-label indication because of the lack of clinical information. However, we were able to determine whether a physician prescribed a medication with no FDA-approved indication for young children to a child between the ages of zero and nine. Fourth, although our data set included data on all U.S. psychiatrists and a random sample of family medicine physicians who prescribed antipsychotics, we did not have data on pediatricians. However, psychiatrists and family medicine physicians together accounted for three quarters (73.6%) of all child antipsychotic prescriptions filled in 2009, and pediatricians accounted for just 9.5% (54).

We found that a large proportion of psychiatrists and family medicine physicians who prescribed antipsychotics wrote prescriptions to young children, although antipsychotic prescribing patterns appeared quite different for this age group compared with patterns for adults. Given the dearth of evidence of antipsychotic efficacy among children and the growing evidence of risks associated with child antipsychotic use, payers and provider organizations could consider interventions that are targeted at physicians who prescribe antipsychotics to young children and highlight the clinical tradeoffs associated with the use of these medications in order to ensure the quality, safety, and value of child mental health treatment.

Hocine Azeni, M.A., provided expert statistical programming.