Off-Label Prescribing of Psychotropic Medication, 2005–2013: An Examination of Potential Influences

We used 2005 and 2013 data from the NDTI to analyze prescriber behavior (6,7). The NDTI obtains diagnostic and treatment data from an ongoing survey conducted by using a two-stage stratified cluster, which is randomly drawn. The survey, conducted by IMS Health, is completed by a panel of 4,140 nationally representative, office-based physicians (N=4,120 in 2005; N=4,140 in 2013). The NDTI panel draws from American Medical Association and America Osteopathic Association listings of physicians practicing in the continental United States. The panel is stratified by region and includes multiple specialties, for example, family and general practitioners and psychiatrists. The NDTI is similar to the National Ambulatory Medical Care Survey (NAMCS), sponsored by the Centers for Disease Control and Prevention, but it has more participants and specifically asks physicians about the condition for which they prescribed a particular medication. Physicians report data quarterly for all patient contacts during two consecutive workdays. Data are collected on approximately 2,760 workdays every month and 8,280 workdays every quarter. The survey uses a confidential logbook in which respondents record information for each drug recommended or issued to patients and the related diagnosis. A patient encounter may generate multiple diagnoses; however, the log provides a direct correspondence between each diagnosis and recommended medication. These recommendations, called “mentions” by the NDTI, may be equal to or greater than actual outpatient prescribing because patients may not fill all prescriptions and because physicians may link a single mention to multiple diagnoses, for example, by prescribing an antidepressant for both depression and anxiety. Results are projected to the national population of office-based physicians (8). Additional information on the sample methodology is provided by Mark (9).

We grouped mentions into three categories of medication (antidepressants, antipsychotics, and anxiolytics) by using the 2014 Uniform System of Classification (10). Sedative hypnotics were excluded from the anxiolytics studied. We used ICD-9-CM (11) codes to categorize diagnoses as psychiatric (codes 290–314) or nonpsychiatric. Diagnoses were grouped with Agency for Healthcare Research and Quality Clinical Classifications Software, a categorization that combines related ICD-9-CM codes (12). We grouped diagnoses into general categories, for example, anxiety disorders and depression. We also listed the ten most commonly identified nonpsychiatric uses for each type of medication in 2013, supplemented with others that were ranked as most common in the 2005 data to allow complete comparison across years, and grouped all others into an “other” category. The percentage of mentions of each drug category per diagnosis for 2005 and 2013 and changes in percentage points between 2005 and 2013 were calculated. Standard errors were calculated by using tables of relative standard errors that accounted for the complex NDTI sampling design, and two-sample z-tests were applied, allowing calculation of p values.

Next, we examined changes in labeling between 2005 and 2013 for drugs in each class by using current and historical labeling data available at Drugs@FDA (13), DailyMed (14), or 2003 and 2005 editions of the Physicians’ Desk Reference (15,16). We examined the extent to which changes in mentions between 2005 and 2013 coincided or conflicted with labeled indications during that time.

Last, a literature scan examined whether changes in the scientific evidence or in major treatment guidelines between 2003 and 2013—apart from labeling changes—might have influenced prescriber behavior.

FDA labeling plays a major role in influencing how providers select medications to prescribe for specific conditions. The addition of new indications for a class of drugs or even the expansion of indications to more drugs in the same class provides an imprimatur for such prescribing and allows pharmaceutical manufacturers to advertise a product for new indications or populations. These data confirm that psychotropic prescribing responds to the addition of new labeled indications. The increased use of antidepressants for anxiety disorders and pain and of antipsychotics for depression, bipolar disorder, and disorders diagnosed in childhood is consistent with a rise in the number of labeling indications as well as with increased information regarding efficacy (23–29). As one example, the increase of 1.5 percentage points in antipsychotic mentions for disorders usually diagnosed in childhood reflects the addition of indications to address irritability associated with autistic disorder and additional indications for tics associated with Tourette’s disorder, prompted by evidence that had accumulated supporting prescribing of antipsychotics for those uses (24–26,30). Antipsychotics were prescribed for these disorders in 2005, before the label changes, but such use increased by 2013, as additional labeling was attained. Antipsychotic use for schizophrenia and other psychotic disorders also increased but by the lesser amount of 1.0 percentage point, accompanying the addition of new medications in the market and new indications for new populations (children and adolescents).

These analyses also confirm that prescribers respond to black box warnings and associated evidence of side effects and risk associated with prescribing medications to particular populations, for particular conditions, or both. The percentage of antidepressant mentions for depression decreased by 5.6 percentage points between 2005 and 2013. This decline was likely associated with black box warnings of the risk of mania and the risk of suicide associated with use of antidepressants among young people. Similarly, use of both antipsychotics and anxiolytics for the treatment of dementia decreased, by 4.6 and .7 percentage points, respectively, reflecting black box warnings, scientific evidence of increased risk of falls and other morbidity among elderly persons (22,31), and policies limiting coverage, such as the exclusion of benzodiazepines from Medicare coverage (21,22,31–34). Our findings related to antipsychotics coincide with those of Driessen and others (33), who found that antipsychotics were increasingly unlikely to be dispensed between 2010 and 2012 for treatment of dementia among elderly Medicare beneficiaries. At no point in the study period were antipsychotics labeled as appropriate for the treatment of dementia-related disorders.

The causes of some changes in prescribing are less obvious. For instance, reasons for the decrease of 2.7 percentage points in antipsychotic mentions for attention-deficit, conduct, and disruptive behavior disorders are ambiguous. As of 2005, chlorpromazine HCL was labeled for severe behavioral problems among children, including short-term treatment of hyperactivity. After 2005, haloperidol also was labeled for this indication, albeit accompanied by distinct warnings that it be used only after other approaches have failed. Countering this labeling addition were published reports indicating a growing concern about increased use of antipsychotics for such disorders, especially involving prescribing by nonpsychiatrists (35,36) and overuse in lieu of other treatments (35–37). These concerns may have outweighed any increase in prescribing associated with the newly labeled indications. Notably, this decrease was accompanied by a nonsignificant increase of 1.4 percentage points in anxiolytic mentions for these disorders, perhaps evidence that one sedating medication had been substituted for another.

This study revealed examples of off-label prescribing that increased between 2005 and 2013 (for example, use of antidepressants for attention-deficit, conduct, and disruptive behavior disorders and use of anxiolytics for bipolar disorder) and examples in which use persisted, albeit at lower rates, despite black box warnings (use of antipsychotics and anxiolytics for dementia). It is important to ask why off-label uses that are contraindicated persist. Tracking on- and off-label prescribing is important to inform the need for education, research, and labeling changes.

One important implication of this work is the need to evaluate barriers and incentives that affect whether pharmaceutical manufacturers apply for labeling changes related to new indications. Companies can file supplemental new drug applications, following earlier approval, to add indications. Recent reviews have found, however, that new drug applications are not commonly used (38). In some instances, labeling for new indications may be appropriate but disincentivized.

Another important consideration is the nature of the current psychiatric diagnostic schema that serves as the basis for labeled indications. These systems (ICD and DSM) are based on psychiatric symptoms, which often are heterogeneous across diagnoses. As the former director of the National Institute of Mental Health (NIMH) pointed out, “The symptom of anxiety, for instance, can represent an endocrine disorder, a psychotic process, a drug response, or one of the currently recognized anxiety disorders.” (39) NIMH is conducting the first steps toward developing a new diagnostic approach that integrates information besides symptoms, but this effort is still in its infancy.

This study had potential limitations. It used mentions of recommended medication as a proxy for actual treatment, which may overstate dispensed medications because patients may not obtain a recommended prescription and may understate dispensed medications because they exclude refills and inpatient prescribing. Therefore, results for certain diagnostic categories or classes of drugs may differ from results that are based on dispensed medications from all settings, such as hospitals, or on data on prescriptions filled. Also, although the NDTI is the only data set that links a medication to a diagnosis, it is possible that participation as a panel member may influence the types of medications mentioned. Although the sample was selected and weighted to be nationally representative, we cannot be certain that the participating physicians were not more cautious in off-label prescribing. We also do not know the response rate for the physicians because that is proprietary information, but we know that the response rate for the NDTI is lower than for government surveys (40). Similarly, the inability to incorporate patient information limited the ability to link changes in drug mentions with age-related changes in labeling. Also, because we did not separate drug mentions by physician type, we cannot report whether the results differed between psychiatrists and other physicians. Last, additional research that studies mentions by specific medication rather than by drug class would allow greater granularity in analyzing off-label prescribing by connecting each drug mention to a particular label.

Despite its limitations, the NDTI is the only data set that links a medication to a diagnosis. The NDTI, claims data, and data from surveys such as the NAMCS all include diagnoses and medication information, but they do not report the information in the same way. In claims or NAMCS data, multiple diagnoses may be listed, and it is impossible to determine the diagnosis for which a medication was prescribed. Similarly, if no psychiatric diagnosis is identified, it is impossible to ascertain the prescriber’s intention. Although the NAMCS could provide similar information, it currently does not. The NDTI, on the other hand, preserves the necessary linkage between medication and diagnosis and provides unique insight into office-based physicians’ perceptions of appropriate pharmacologic treatment. Similarly, although claims data capture only prescriptions reimbursed by the pertinent payer, NDTI data include information about all prescriptions, regardless of insurer, including prescriptions paid for out of pocket. Rather, the data capture the prescription the provider would “write” if reimbursement and patient preference were not considerations. In that sense, they capture the provider’s vision of what should be prescribed rather than what can be prescribed.

These findings highlight that labels are an important but imperfect guide for physicians to make prescribing decisions. This may be particularly true in psychiatry, given that the diagnostic schema that forms the basis for labels is highly imprecise. Tracking prescribing for labeled and off-label conditions is a useful tool for illuminating whether new labeled indications may be appropriate, whether research on new indications and contraindicated indications is needed, and whether physicians may benefit from additional education about the application of particular medications for particular clinical scenarios.

The authors acknowledge the assistance and advice of Catherine A. Fullerton, M.D., M.P.H, Linda Lee, Ph.D., and Ali Hashmi, M.S., of Truven Health Analytics, L.L.C., in manuscript preparation as well as the contribution of Mitchell Berger, M.P.H., and others at SAMHSA.