Several consensus guidelines for the use of antipsychotic drugs have been published (
1–
4). These guidelines generally recommend against prescribing high doses of antipsychotics (
1,
2,
5). Higher doses, even within the recommended ranges, are associated with greater risks of side effects, including (depending on the drug) extrapyramidal symptoms, weight gain and associated metabolic effects (
6), prolactin elevation and associated sexual side effects (
7), poorer cognitive functioning (
8), and QTc prolongation (
1,
9). Clozapine is recommended for patients with treatment-resistant schizophrenia (
1,
10).
Guidelines also recommend against antipsychotic polypharmacy, defined as coprescription of two or more antipsychotic drugs, except over a short period of switching antipsychotics (
1,
2,
5). Recent evidence suggests that in cases of refractory schizophrenia, augmentation of treatment with clozapine or with aripiprazole to mitigate metabolic side effects or both approaches may be helpful in some clinical situations (
10,
11). Two recent large studies have called into question the long-held belief that antipsychotic polypharmacy increases mortality (
10,
12–
14). Nonetheless, several potential harms of antipsychotic polypharmacy remain well established. Antipsychotic polypharmacy tends to increase total dose and thus concomitant side effects. Less obviously, interactions among antipsychotics, mediated through cytochrome p450 enzymes, can cause unexpected increases in peripheral drug concentrations, with a concomitant increase in the incidence and severity of side effects (
10,
15). In addition, antipsychotic polypharmacy tends to be more costly than monotherapy (
16). Overall, antipsychotic polypharmacy presents more risks and disadvantages than benefits in most circumstances (
10,
17,
18).
Yet numerous surveys indicate that both high-dose prescribing and antipsychotic polypharmacy are common (
10,
19–
22). Limited evidence suggests that some prescribers contribute more than others to relatively high rates of high-dose prescribing and polypharmacy independently of patient characteristics (
23,
24). Physician specialty and number of years in practice as well as institutional specialty, academic emphasis, or unit specialty have been associated with prescribing of high-dose antipsychotics and antipsychotic polypharmacy independently of patient characteristics (
23,
25,
26).
The purpose of our study was to examine the extent of variation in high-dose and polypharmacy prescribing of antipsychotics at the physician and hospital levels over a sufficiently long period to rule out any transitory prescribing patterns. To the extent that some practitioners and treatment settings tend to systematically favor high-dose prescribing and antipsychotic polypharmacy, even after adjustment for patient characteristics, targeted interventions may be warranted.
Methods
Methods are briefly described below. [Details are provided in an online
data supplement to this article.] Ethics approval for the study was obtained from the Research Ethics Board of the Douglas Mental Health University Institute.
Study setting
This study was carried out with data from the Canadian province of Québec (population 8 million). Psychiatric care is sectorized in Montreal, the province’s largest city (population of Montreal health region 1.8 million). As a result, although more severely ill patients tend to seek care from the province’s psychiatric hospitals, people with mental illness normally receive care from the hospital that is geographically closest to them. All senior citizens (age ≥65), all public assistance or disability benefit recipients, and all other individuals not eligible for coverage by a private plan (about 30% of the total population) have extensive prescription coverage through a public plan. All provincial residents also have nearly complete public coverage for physician services and complete coverage for hospitalizations.
Analytic population
Linking provincial databases recording physician claims as well as hospitalizations yielded 21,535 individuals who were at least 18 years old at the end of 2004, and for whom at least two-thirds of mental illness diagnoses on either physician claims or hospitalization records were for schizophrenia (
ICD-9-CM codes 295.0–295.9) during 2003 or 2004. Via further linkage with the drug claims database, we estimated that 56% of this group, or 12,150, had antipsychotic prescriptions covering at least 11 months (335 days) during 2004 (
15,
27).
Definition of a high dose
Estimated daily dosages of antipsychotics were converted into chlorpromazine equivalents (CPZE), a measure of the relative potency of antipsychotic drugs. Schizophrenia Patient Outcomes Research Team guidelines recommend daily antipsychotic doses between 300 and 1,000 mg CPZE for first-generation antipsychotics (
1). Accordingly, we defined a high dose for these drugs as one exceeding 1,000 mg CPZE (
28). For second-generation antipsychotics, this threshold was adjusted downward to 808 mg CPZE, corresponding to the 85th percentile of the distribution of average daily CPZE doses among patients taking only first-generation antipsychotics that we observed in 2004. We used, conservatively, a similar procedure to define a high dose for long-acting injectable medications: the 85th percentile of dosages for these drugs alone or in combination with oral medication corresponds to 1,556 mg CPZE.
Patient and physician characteristics
Patient characteristics included sex, age, and public assistance and disability benefits status. Long-term benefits enrollees typically are more ill than others. A binary variable indicating whether the physician had prescribed clozapine in 2004 was used as a proxy for average severity of illness among physicians’ patients. Physicians also were linked to a database containing specialty, university where they received their medical training, and year of graduation.
Linking of patients to physicians
Each prescription record included a physician identifier. For each patient during 2004, a physician was determined to be either the sole prescriber of psychotropic medications or one of two or more prescribers. In all, 7,430 patients (61.2% of our population) had a unique physician prescriber of psychotropic medications. Seventy-two percent of the physicians in the data set (1,581 of 2,191) had at least one patient for whom they were the sole prescriber.
Physician patient load
A physician’s (exclusive) patient load was calculated as the total number of patients to whom the physician prescribed antipsychotic medications and for whom that physician was the sole prescriber of psychotropic medications. The distribution of the number of patients per physician was divided into tertiles: small (one to five patients), medium (six to 12), and large (≥13).
Linking of patients to hospitals
Patients were linked to hospitals in one of two ways. First, if they had a sole prescribing physician and that physician was linked to a hospital, they were linked to that physician’s hospital. Physicians were linked to a hospital if 80% or more of their patients for whom they were the sole prescriber were hospitalized with a principal psychiatric diagnosis at the same hospital. (Patients for whom a physician was their sole prescriber could be linked unambiguously to that physician.) Second, if patients had more than one prescribing physician, but all of them were associated with the same hospital, these patients were assigned to that hospital.
This linkage method yielded 7,810 patients (64% of the 12,150 continuously on antipsychotics) who could be assigned to a hospital. Based on the tertiles of the number of patients linked to each hospital, small hospitals were defined as having 54 patients or fewer, medium hospitals 55–146 patients, and large ones 147 patients or more.
Hospital characteristics
The data were released to us in a manner precluding precise identification of hospitals. However, from region and number of patients, we could classify hospitals as urban nonpsychiatric, rural nonpsychiatric, or psychiatric.
Data analysis
We first calculated the proportions of patients receiving high doses or polypharmacy of antipsychotics, and the proportion receiving both. We also calculated, among those on polypharmacy, the proportion to whom clozapine was one of the drugs prescribed. All computations were carried out with Stata, version 12.
Using Stata’s GLLAMM estimation procedure, we then estimated both a simple logistic regression model and a generalized linear mixed model, with receipt of high-dose antipsychotics as the outcome. The latter model enabled us to calculate the proportion of the remaining unexplained variance that could be attributed to systematic, but unobserved, patient-, physician-, and hospital-level factors. For computational reasons, the GLLAMM procedure had to be estimated with a sample that excluded patients who were the sole patient of a physician. This sample included 11,364 patients out of 12,150 (93.5%). To see to what extent the GLLAMM estimates obtained on the reduced sample differed from the original set of logistic regression estimates because of the difference in estimation method as opposed to the difference in sample, we also estimated a simple logistic regression model with the reduced sample.
We then calculated both the raw proportion of patients on a high dose for each physician and an adjusted proportion (calculated using logistic regression estimates) that removed the effect of observed patient characteristics. The observed and adjusted proportions of patients on a high dose were stratified by physician patient load and plotted. Analogous calculations by hospital were carried out with the proportion of patients on high doses.
These analyses were repeated with receipt of antipsychotic polypharmacy as the outcome variable.
Sensitivity analyses
All calculations were repeated with alternative thresholds of two months or more and six months or more, for receiving a high dose and for receiving polypharmacy.
Results
Overall, 11.9% (1,447 of 12,150) of people who took antipsychotics were on consistently high doses, 10.4% (N=1,266) were on antipsychotic polypharmacy, and 3.7% (N=451) were on both high doses and antipsychotic polypharmacy. Of the 1,266 patients on antipsychotic polypharmacy, 16.7% (N=211) were prescribed clozapine.
Table 1 shows patient, physician, and hospital characteristics that we were able to associate with particular patients.
Table 2 presents estimates obtained from the three models for predicting whether a patient was on a high dose. In all models, males were more likely to be kept on high doses, and the probability of being on a high dose showed a ∩-shaped relationship with age, with the peak occurring in patients’ fifties. Patients never or sometimes receiving public assistance were less likely than those 65 or over to be on high doses. Patients assigned to a hospital were more likely to be on a high dose, meaning (because we controlled for having an assigned physician) that having an assigned physician whose activities were concentrated in one hospital, or more than one prescriber who was assigned to the same hospital, was associated with being prescribed a high dose. Patients whose physicians had prescribed clozapine to any patient in 2004 were significantly more likely to be on a high dose.
The logistic regression models showed an association between physician’s university of graduation, and being assigned to a psychiatric hospital, and likelihood of being on a high dose. A systematic propensity for some physicians to prescribe high doses accounted for 10.9% of the remaining unexplained variance and for physicians to prescribe high doses in a common way across some hospitals accounted for 3.0%. The remainder must, logically, be attributed to unobserved patient-level factors.
Results for antipsychotic polypharmacy were broadly similar (
Table 3). Notable differences included the following: being male and receiving public assistance were not significantly associated with the probability of being on antipsychotic polypharmacy; having an assigned physician (meaning a sole prescriber) greatly increased the likelihood of being on polypharmacy, whereas it appeared to have no such effect on the likelihood of being on a high dose. Being assigned to a psychiatric hospital was significantly associated with being on long-term polypharmacy only in the final model. A systematic propensity for some physicians to prescribe high doses accounted for 9.7% of the remaining unexplained variance, and for physicians to prescribe high doses in a common way across some hospitals, for 6.2%.
[A figure in the online
data supplement shows the distribution of observed versus adjusted proportions of physicians’ patients who were receiving high doses side by side with those receiving polypharmacy. A second figure shows the distributions for hospitals. Both figures show considerable variability in the average proportion of patients on high doses or on polypharmacy.]
Finally, sensitivity analyses indicated that the percentages of patients on polypharmacy for two and six months (in our sample of 12,150) were 31.1% (N=3,777) and 19.7% (N=2,395), versus 22.5% (N=2,739) and 18.4% (N=2,238) for high doses. Results (not shown) also indicated that the regression model coefficients were broadly similar to the alternative duration thresholds, with two significant differences: first, having been assigned a physician was not associated with a higher probability of being on antipsychotic polypharmacy for two months (for the full sample, odds ratio=.79, 95% confidence interval=.67–.94), whereas it was for six and 11 months, and second, physician’s university of graduation was not significantly associated with the probability of being on a high dose or polypharmacy for two months or six months. Physician-level unobserved factors accounted for a declining proportion of the unexplained variance in high-dose prescribing as threshold declined, decreasing from 10.9% to 6.2% at the two-month threshold. The direction of change was reversed for hospital-level factors: the proportion increased from 3.0% to 4.1%. In the case of antipsychotic polypharmacy, as the threshold duration declined, the proportions of the unexplained variance accounted for by physician-level and hospital-level factors both declined, from 9.7% to 6.0% at the two-month threshold for physician-level factors and from 6.2% to 1.8% for hospital-level factors.
Discussion
The percentage of people with schizophrenia in Québec who were taking antipsychotics regularly and whose illness was maintained long-term with high doses of antipsychotics (11.9%) is difficult to compare with that from other studies. To our knowledge, prior studies have examined high-dose prescribing over periods of only one month or less in populations that may not have been taking antipsychotics consistently (
20,
23,
24,
26,
28–
30). These studies have reported rates from 10.1% in a large U.S. Department of Veterans Affairs population (
26) to 30.2% in a Japanese sample (
20). As for antipsychotic polypharmacy, although most studies have reported rates of polypharmacy over periods of one month or less of 50% or more (
20,
24,
28,
31), studies with follow-up periods of a few months have tended to report lower rates, typically ranging from 20% to 30% (
32–
34). In one study of disabled California Medicaid beneficiaries with schizophrenia in 2004, about 2% of the sample was on second-generation antipsychotic polypharmacy for the entire year (
35). In our study, 5.9% of our sample (N=1,266 of 21,535) identified as having schizophrenia and receiving at least some antipsychotic treatment were on some type of antipsychotic polypharmacy for 11 months or more. The figures are not entirely comparable, although they suggest that long-term polypharmacy may have been more common in Québec than in California at that time. In summary, a significant minority of patients (N=2,262) were maintained on long-term high doses, polypharmacy, or both when guidelines call for neither. They represented 18.6% of those consistently on antipsychotics and 10.5% of those identified as having schizophrenia.
Have clinical guidelines not caught up with clinical experience? Coprescription with clozapine for treatment of refractory schizophrenia may justify antipsychotic polypharmacy (
10,
11), but in only 16.7% of cases in our study was one of the drugs clozapine. A number of surveys have sought to determine why physicians prescribe antipsychotic polypharmacy or high doses. Antipsychotic polypharmacy in community settings can result from a desire to treat persistent positive symptoms or to alleviate sleep disturbance (
36) or from finding that a combination arrived at as part of a switching process seems to work well (
36,
37). Prescribing high doses can also be the result of antipsychotic polypharmacy (
31,
33). In addition, factors not related to patients’ current behavior, such as a history of aggression and a longer history of antipsychotic prescription, have been implicated (
23).
Our results suggest that although such explanations undoubtedly account for individual cases of high-dose prescribing and antipsychotic polypharmacy, they do not account for the finding that prescribers vary significantly in their propensity to prescribe high doses or antipsychotic polypharmacy. This finding is consistent with that reported in other, smaller studies (
24,
38).
Reducing the proportion of patients on antipsychotic polypharmacy, and thereby at least to some extent on high doses of antipsychotics, may be clinically feasible. In one trial, two-thirds of adults with schizophrenia who were switched to monotherapy did so successfully, although it appeared preferable to leave open the option to return to polypharmacy (
39).
Audit and feedback mechanisms may also be helpful. These mechanisms, which have shown some, albeit limited, effectiveness, could target physicians who seem to have the greatest propensity to prescribe high doses or antipsychotic polytherapy (
40,
41). More broadly, these mechanisms could target entire hospital departments in cases where the data suggest that physicians in these departments tend as a group to systematically keep a greater proportion of their patients on high doses or antipsychotic polypharmacy (
42,
43).
Such audit and feedback mechanisms may need to address high-dose prescribing and polypharmacy separately: among the 435 physicians who were the sole prescribers for four or more patients with schizophrenia, 119 (27%) were in the highest quartile of the proportion of patients on a high dose, on polypharmacy, or on both. Of these 119 physicians, 50 (42%) were in the upper quartile of both; thus more than half of the physicians were in the highest quartile in terms of either high-dose prescribing or antipsychotic polypharmacy, but not both at once.
In addition, because both high-dose prescribing and antipsychotic polypharmacy can result from a reluctance to prescribe clozapine (
42), mechanisms to facilitate the prescribing of clozapine may be helpful (
15,
44). Finally, residency training could be reviewed in light of our observation that university of graduation was associated with the propensity to maintain patients on polypharmacy.
This study had significant strengths. Unlike most previous studies, it identified high-dose prescribing and antipsychotic polypharmacy over a total of 11 months out of a calendar year and thus reflects long-term prescribing patterns, which are of particular concern clinically. In addition, the observations were based on a large number of people with schizophrenia, identified with administrative data. This study is, to our knowledge, the first population-based study of this issue to characterize prescribing patterns at the physician level and thus observe the extent of variability in prescribing patterns across physicians, with controls for some differences in patient characteristics. Finally, this is also the first study, to our knowledge, to use generalized linear mixed modeling to estimate the percentages of unexplained variation in physicians’ propensity to prescribe high doses and polypharmacy that can be accounted for by unobserved patient-level, physician-level, and hospital-level factors. These methodological refinements allowed us to draw more precise inferences concerning the most efficient types of measures that could be implemented to bring physicians’ prescribing patterns more uniformly close to practice guidelines.
The study also had some limitations. First, the use of administrative data limited our ability to adjust for patient characteristics and thus might account for some of the apparent variability in prescribing behavior across physicians and hospitals. Second, the data linkage procedures resulted in some loss of patients, physicians, and hospitals from the analyses. This may reduce the generalizability of our findings, although we kept data for the physicians of most interest for our analyses. Third, the proportion of the total Canadian population with schizophrenia has been estimated to exceed .6% (
40), whereas we were able to identify only .3% from administrative data. Fourth, the data were from 2004, before the introduction of some new antipsychotics, which may have affected practice patterns to some degree. However, no policy measures have been introduced in Québec since then that are likely to have affected our central finding of material and unexplained variability in these prescribing patterns across physicians and hospitals. Fifth, our results reflect the behavior of physicians during a particular year, and it is possible that different physicians would have emerged as tending to prescribe high doses or polypharmacy in another year. Given that the phenomenon reported here is consistent prescribing over essentially an entire year, it seems unlikely that there would be much change in the proportions of a physician’s patients on high doses or polypharmacy from one year to the next. Further research could shed more light on the stability over time of such prescribing patterns.
Conclusions
Taken together, these findings suggest that high-dose prescribing and antipsychotic polypharmacy cannot be explained simply as responses by individual physicians to particularly difficult clinical situations. Rather, they appear to result, to a significant extent, from a minority of physicians’ and of hospital departments’ more atypical prescribing practices. Given the harms associated with both high-dose prescribing and antipsychotic polypharmacy, measures to reduce such prescribing are urgently needed.
Acknowledgments and disclosures
This study was funded by grant 6659 from the Fonds de la recherche Québec–Santé (Québec Health Research Fund). The authors also thank the Régie de l’Assurance-Maladie du Québec (Québec Health Insurance Board) for providing data.
Dr. Malla has received investigator-initiated research grants and honoraria for serving on advisory boards, for consulting, and for conference presentations that were sponsored by the pharmaceutical industry and included Janssen, Pfizer, Otsuka, Lundbeck, and Bristol-Myers Squibb. The other authors report no competing interests.