This study examined the impact of prior authorization formularies on the likelihood that patients with schizophrenia will be arrested and incarcerated.
To examine the impact of Medicaid prior authorization for atypical antipsychotics on the prevalence of schizophrenia among the prison population.
We collected drug-level information on prior authorization restrictions from Medicaid programs in 30 states to determine which states had prior authorization requirements before 2004. We linked the regulatory data to a survey of prison inmates conducted in 2004.
We used a sample of 16,844 inmates from a nationally representative survey and analyzed the data using cross-sectional regression. To capture the impact of prior authorization, we estimated 2 models: the first included an indicator variable for states requiring prior authorization, and a second model used per capita atypical usage.
Evidence indicated that prior authorization restrictions on atypical antipsychotics are associated with an increase in the odds of a schizophrenic resident being imprisoned in a state. State-level prior authorization requirements for atypical antipsychotics are associated with a 2.7% increase in the likelihood that an imprisoned inmate displays psychotic symptoms, and a 1.25 increase in the likelihood that an inmate was previously diagnosed with schizophrenia by a physician. Higher state-level atypical prescriptions per capita are also associated with lower likelihood of psychotic symptoms and of prior schizophrenia diagnosis among prisoners.
Prior authorization requirements for atypical antipsychotics, which are designed to reduce healthcare costs, are associated with greater prevalence of mental illness within the criminal justice system.This association raises important questions about whether increased costs to the criminal justice system might mitigate or offset prescription drug savings created by prior authorization requirements.
Am J Manag Care. 2014;20(7):577-586We examined the impact of prior authorization formularies on the likelihood that patients with schizophrenia will be arrested and incarcerated.
These findings suggest that prior authorization requirements for atypical antipsychotics, which are designed to reduce healthcare costs, can result in increased expenditures for the criminal justice system. Given the high cost of incarceration, these increased costs to the criminal justice system could easily offset any savings created by prior authorization requirements.
Increasingly, public and private healthcare plans are imposing prior authorization requirements to manage drug spending. Prior authorization plans establish a preferred drugs list (PDL) and require authorization before covering drugs not on the PDL. Such programs are designed to reduce costs by steering utilization toward lower-cost medications, but these policies may also lead to poorer drug adherence.1 Poor adherence to or discontinuation of antipsychotic drugs among patients with schizophrenia is likely to cause acute psychotic episodes2-4 and often results in contact with law enforcement officers due to threatening behaviors brought on by active symptoms, leading to arrest and incarceration.5
While schizophrenia affects only about 1% of the US population,6 it is difficult and expensive to treat and can have a devastating impact if not well controlled.7 The effects of losing continuous effective exposure to medication due to nonadherence are immediate and evident. A study on the temporal relationship between medication nonadherence and hospitalization risk for individuals with schizophrenia found that individuals in the first 10 days following a missed prescription refill had a greater than 50% increase in the risk of mental health hospitalization and a 77% increase in the risk of schizophrenia-specific hospitalization.8
As a result of their behavior, mentally ill people are more likely to be arrested than other people who are stopped by police, and more likely to be subsequently convicted and incarcerated.9 Enforcement officials may not be sufficiently trained to recognize mentally ill patients, or they may feel obliged to bring the schizophrenic offender to the criminal justice system when the healthcare system is unwilling or unable to accept a violent patient. Prison systems are often poorly equipped to treat the mentally ill, and, when released these prisoners are more likely to recidivate,10 which promotes the vicious cycle of mentally ill criminals entering the criminal justice system, suffering further mental health deterioration in prison, and upon release ending up with a significant likelihood of re-arrest due to poorly controlled mental illness.
Evidence also shows that the mentally ill are more costly to incarcerate than those without mental illness and that other types of rehabilitation for the mentally ill may reduce crime at costs lower than those associated with incarceration.11 Over the past several decades, pharmaceutical advancements in the treatment of schizophrenia have been shown to reduce the likelihood of episodes that start the sequence of events leading to incarceration of patients. The second-generation antipsychotics known as atypicals were introduced in the 1990s. These have largely replaced older antipsychotics, because they are associated with lower rates of relapse. About 30% to 40% of patients relapsed with first-generation drugs (relative to 80% without treatment); for second-generation drugs, relapse rates fell to about 25% to 29%.12-14 Furthermore, many patients respond to only 1 drug, so that increased treatment options increase the likelihood of successful treatment.15 The lower rates of relapse seen with atypicals may lead to fewer incidents with law enforcement and reduced incarceration of schizophrenia patients.
The evidence suggests that atypicals improve outcomes for certain patients with schizophrenia. The Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study found that perphenazine (Trilafon), an older typical, worked about as well as several newer atypicals for the average patient in the trial.16,17 Subsequent research on the CATIE data clarified that a subset of patients in CATIE strictly benefited from atypical usage, and that providing typicals to all patients would have reduced overall health.18 Taken together, this literature suggests that atypicals represent a valuable treatment option for some patients with schizophrenia.
In this study, we examine the association between Medicaid policy and 2 key outcomes-—namely, utilization of antipsychotics and imprisonment of patients with schizophrenia. We compared the prevalence of patients in prison among states that did or did not institute prior authorization programs for atypical antipsychotics. We also compared prevalence across states that have different levels of atypical antipsychotic utilization. Because of the association between prior authorization regulations and increased risk of discontinuing atypical treatment, and because of the potential link between untreated schizophrenia and incarceration, we hypothesized that prior authorization rules will increase the likelihood that a schizophrenia patient is incarcerated.
Ideally, we would like to have measured the percentage of each state’s population with schizophrenia that is incarcerated, and then examined how this percentage is associated with state prior authorization policy. In practice, we were able to obtain only the percentage of the incarcerated population who were diagnosed as having schizophrenia in each state. If it is assumed that Medicaid pharmacy regulations have no impact on the overall prevalence of schizophrenia in a state, and that they have negligible impact on the rate at which nonschizophrenics are imprisoned, then the impact of pharmacy regulations on the prevalence of schizophrenia in prisons will be comparable with the impact of these regulations on the rate of imprisonment of patients with schizophrenia.
A simplified summary of the underlying process illustrates how restrictive pharmacy policies in Medicaid might lead to a higher prevalence of schizophrenia in prison:
Stage 1: A large majority of patients with schizophrenia are covered by Medicaid.
Stage 2: The state’s restrictive prior authorization policy leads to discontinued, reduced, or inappropriate medication.
Stage 3: Suboptimal medication leads to uncontrolled symptoms.
Stage 4: Active symptoms or deteriorated economic conditions that result from uncontrolled symptoms lead to encounters with law enforcement and arrest.
Stage 5: Repeated or serious arrests lead to imprisonment.
Our hypothesis is that Medicaid prior authorization regulations limit access to atypicals, and that this increases the likelihood that a patient will progress down the list from stage 2 to stage 5.
Our analysis made use of the 2004 Survey of Inmates in State Correctional Facilities (SISCF) and the 2004 Survey of Inmates in Federal Correctional Facilities (SIFCF) conducted for the Bureau of Justice Statistics by the Bureau of the Census; these surveys have been used in previous research.19 They provide nationally representative data on inmates held in state and federal prisons obtained through personal interviews of more than 18,000 inmates in about 300 prisons. These data contain individual-level information on inmates’ mental health conditions and various personal characteristics. A benefit of the SISCF data is that they also contain state indicators, which allow us to match the data to our original survey of antipsychotic coverage.
Using a mailed survey that was conducted in 2009, we collected information from 30 state Medicaid programs on their utilization review policies for atypical antipsychotics over the period 1999 to 2008.20 This survey asked whether prior authorization policies applied to a list of drugs identified by US brand name. To supplement the survey information, we examined Medicaid pharmacy program websites for relevant documents and contacted Medicaid program personnel. This allowed us to identify the 4 states (AK, CA, MA, and NY) that had prior authorization for all atypicals before 2003, along with 24 states that we can determine had no prior authorization before the 2004 SISCF survey. For the remaining states, we are unable to determine the prior authorization policy in 2003, because information on the timing of prior authorization implementation is unclear or unavailable from the mailed survey or websites. (See for a list of states by policy status.)
State-level utilization data were obtained from the State Drug Utilization Data Files available from CMS. These data, which include the number of prescriptions filled for each drug by each state Medicaid program by calendar quarter, were obtained for 2003 atypical antipsychotics. State-level Medicaid enrollments as of June of each year were obtained from the Kaiser Family Foundation’s State Health Facts database.
Prevalence in the General Population
We used prevalence of serious psychological distress (SPD) estimated from the 2003-2004 National Surveys on Drug Use and Health, produced by the Substance Abuse and Mental Health Services Administration (SAMHSA), as a proxy for the prevalence of schizophrenia in the general population. Serious psychological distress was measured using the K6 screening instrument for nonspecific psychological distress. In 2003-2004, SPD was noted in about 10% of the population 18 years or older. If we can assume that the percentage of schizophrenia in SPD does not vary by status of prior authorization or level of atypical utilization, then such an approximation does not affect interpretation of our results.
We estimated the probability that an inmate is screened as positive for psychotic symptoms, and whether an inmate reports a prior schizophrenia diagnosis, using a linear probability model. In addition to an indicator variable that equals 1 if the state in which the inmate is incarcerated has a prior authorization rule for atypical antipsychotics, we included the prevalence of schizophrenia in the general population in the state and several individual inmate characteristics found in .
eAppendix Tables 1
As documented in Table 1, the probability that a prisoner was previously diagnosed with schizophrenia is relatively low, and this suggests the linear probability modeling approach. Logistic regression is known to perform poorly in binary dependent variable models where positive outcomes are statistically rare; in contrast, linear probability model performance is invariant to the mean of the dependent variable.21 In any event, however, the association between prior authorization (or atypical usage) and the prevalence of mental illness within prisons is robust and statistically significant across linear probability, logistic regression, and probit models. Therefore, our results are not primarily dependent on functional form. All our modeling results are presented for the logistic regression model in and. The results are qualitatively similar.
The outcome variable is an indicator of whether an inmate has screened positive for psychotic symptoms. We utilize 2 questions from the survey. In the first, the survey asked inmates whether during the past year they had seen or heard things that other people said were not there, felt that other people were able to read or control their mind, or felt that someone other than the corrections staff had been spying on or plotting against the inmate. Inmates who answered yes to any of these questions were screened as positive for psychotic symptoms in our analysis. This question measures the presence of psychotic symptoms. Note that it is broader than schizophrenia. Our second measure focuses on a much narrower definition, based on a question that asks the inmate if they have ever been diagnosed by a healthcare provider with schizophrenia or a psychotic disorder. This definition will undercount those inmates who are unwilling or unable to identify themselves as schizophrenic but has the advantage of not resulting from self-diagnosis. As shown in Table 1, 2840 inmates screened positive using the more inclusive definition while 775 reported a prior diagnosis. Significantly, all those with a prior diagnosis also tested positive for the presence of psychotic symptoms.
Two alternative measures of an inmate’s access to atypicals were used. The first is whether the state’s Medicaid program has a prior approval requirement in its formulary, and the second is the Medicaid utilization level of atypical antipsychotics in an inmate’s resident state. Because acute symptoms of schizophrenia can occur after only a few days of missing medication, the lapse between regulation and the effects on imprisonment of schizophrenia patients is likely to be relatively short.
The analysis also included a group of control variables that are commonly suspected to contribute to the onset of schizophrenia: age, gender, race, whether born in the United States, whether completed high school, whether lived with parents when growing up, whether sexually abused as a child, whether caretaker had substance abuse problems, and whether physically abused before admission to prison.
Definition of Subsamples
Along with the full sample of inmates, results were estimated separately on the basis of the type of crime for which the inmate had been convicted. These conviction subgroups include those convicted of 1) violent, 2) drug-related, or 3) nonviolent and nondrug offenses. Note that an inmate convicted of both violent crimes and drug crimes was placed in the violent crime group. Some inmates were excluded from subpopulation analysis because available information regarding their offenses was insufficient.
Characteristics of Inmates in States With and Without Prior Authorization
tabulates characteristics of prison inmates in states with and without prior authorization policy for atypicals, as well as state-level characteristics. All statistics were weighted by the inmate population.
A total of 16,844 surveyed inmates from 48 states were included in either regulation analysis or utilization analysis, representing 1,254,589 inmates nationally. Of the surveyed inmates, 3205 came from 4 states (AK, CA, MA, and NY) that instituted prior authorization for atypicals in 2003; 10,375 came from 24 states that had no prior authorization for atypicals in 2003; and the remaining 3264 came from states for which regulation information was unknown.
About a quarter of the inmates resided in states that instituted prior authorization for atypicals in 2003. In these states, the prevalence of serious mental illness in the general population is lower (10.9% vs 11.7%) but the percentage of inmates who screened positive for psychotic symptoms in prison is slightly higher (17.8% vs 17.0% when screened by psychotic symptoms and 5.05% vs 4.35% when screened by prior schizophrenia diagnosis). Per capita Medicaid prescriptions for atypicals are also lower (0.43 vs 0.51) compared with those in states without such regulations.
Impact of Prior Authorization on Imprisonment of Those With Schizophrenia
In Table 2, we observed a slightly higher percentage of inmates with prior diagnoses of schizophrenia, and with symptoms of psychosis, in states that instituted prior authorization despite a lower percentage of serious mental illness in the general population and a better educated prison population in these states.
Table 3, Panel A
presents the results of a linear probability model of the presence of psychotic symptoms. The dependent variable indicates whether the inmate has been screened and determined to display symptoms of psychosis. The primary explanatory variable of interest in Table 3 is an indicator of whether that inmate’s state of residence instituted prior authorization for atypicals in 2003. The explanatory variable of interest in Table 4 is the per enrollee prescription of atypicals in that inmate’s state Medicaid program in 2003. Column 1 estimates the linear probability model for the entire inmate population, and columns 2 to 4 limit the analysis to subsamples of inmates with violent crime, with drug offenses, or with nonviolent, nondrug offenses. Here and elsewhere, the coefficients on prior authorization are statistically significant in the “all inmates” group for both probit and logistic regression models, in addition to the linear probability models reported.
Table 3, Panel A shows that the restrictive pharmacy policy variable is associated with higher likelihood that an imprisoned individual displays psychotic symptoms. The estimate shows that the probability an inmate displays symptoms increases by 2.7 percentage points 0.009- 0.046] in a state with a prior authorization requirement for atypicals relative to a state without such a requirement. This demonstrates that prior authorization is associated with higher rates of incarceration. The point estimate for the violent crime population is very close to that estimated for the entire population; it is slightly larger for the nondrug, nonviolent population, but the difference between the 2 estimates is not statistically significant. The estimated rates for nonviolent crime are not statistically different in pre-approval and control states.
The association between prior authorization and psychotic symptoms is somewhat higher among the inmate population with drug offenses, although this difference is not robust across alternative approaches to measuring the outcome of interest.
All personal experience factors included in these models proved to be significant and influential: growing up with at least 1 parent and completing high school are associated with a reduced risk of psychotic symptoms, but parents’ substance abuse, sexual abuse in childhood, and physical abuse are all associated with an increased risk of psychotic symptoms. Relative to white inmates, African American inmates have an increased risk of symptoms in prison, as do other minorities, but age and gender did not prove significant.
In Table 3, Panel B, we re-estimate the model using the physician diagnosis-based measure. Consistent with the fact that the prevalence of physician diagnosis is lower, the estimated associations are also smaller in magnitude but still statistically significant and quite meaningful on a percentage basis. For example, the likelihood that an inmate has been diagnosed with schizophrenia increases by 1.2 percentage points [CI, 0.002-0.023]. The number rises to 2 percentage points [CI, 0.000-0.039] for violent crime and 2.2 percentage points [CI, 0.000-0.045] for nonviolent non-drug offenses. The latter results are statistically significant at the 10% level, but insignificant at the 5% level. (The lower limits of the 2 confidence intervals are rounded up to zero.) Recall that the mean prevalence of previous physician diagnosis was around 5%, suggesting that, for example, a 1.2% increase corresponds to roughly a 24% increase.
Impact of Atypical Utilization on Imprisonment of Those With Schizophrenia
We have examined the most common method of reducing atypical use: requiring prior approval. For confirmation of these results, we examined whether broader use of atypicals affects the prevalence of diagnosed schizophrenia and psychotic symptoms in the state’s prison population. Our estimation model is similar to those used in Table 3, although we removed our prior approval control and included the number of atypicals per capita for the state’s Medicaid population. This approach increased our sample size, because we did not need to identify the specific state’s regulation in 2003 and could include all states in the analysis. Table 4, Panel A, shows that a decrease in the variable measuring utilization of atypical antipsychotics is associated with higher likelihood of psychotic symptoms among the imprisoned population. We find a similar impact when we confine the sample to inmates with nonviolent drug offenses. The effects of the control variables were similar to those seen in the earlier analysis studying prior authorization rules. As shown in Table 4, Panel B, the results for physician-diagnosed schizophrenia are similar, and (as before) smaller in magnitude. Higher usage of atypicals is associated with higher likelihood of prior physician diagnosis of schizophrenia in the full sample, among violent offenders, and among those arrested for nonviolent, nondrug crimes. The coefficients on atypical usage are also statistically significant in the “all inmates” group for both probit and logistic regression models, in addition to the linear probability models reported.
There are several limitations to the study, mostly resulting from the cross-sectional nature of the survey. Although we have controlled for a number of individual characteristics, it is still possible that the states with prior authorization requirements differ in some systematic way from those without in their willingness to incarcerate schizophrenic residents. It is relevant to note here that states with prior authorization requirements displayed lower schizophrenia prevalence in the general population, but higher prevalence within the incarcerated population. Moreover, states with lower atypical usage rates also displayed higher prevalence of psychotic symptoms and diagnosed schizophrenia within the incarcerated population; this result is identified across a broader set of states than just the ones with prior authorization. It is not obvious what causal mechanisms would account for all these results, although the cross-sectional data cannot rule out such mechanisms directly.
The second limitation is that our 2 survey-based screens rely either on self-reported symptoms that include psychoses other than schizophrenia, or on recollection of a physician diagnosis. The former may be overinclusive, while the latter may exclude some undiagnosed cases or patients who cannot accurately recall a diagnosis. It is nonetheless encouraging that results are reasonably robust across the measures.
Third, we do not directly observe the mechanism that runs between prior authorization and incarceration. Ideally, one would observe that prior authorization leads to lower medication use among patients with disease, and this is then associated with higher rates of incarceration. Because we do not observe in any data the change in prisoners’ medication use before and after the adoption of prior authorization, it is not possible to test for this mechanism. As a result, our results are suggestive of this mechanism, but do not demonstrate it explicitly.
Finally, we note that there are a number of unmeasured factors that affect rates of imprisonment, above and beyond the formulary policies studied. This is demonstrated by the relatively low R2 values in our estimated models.
We found that prior authorization requirements for the use of atypical antipsychotics in states’ Medicaid coverage are associated with an increased likelihood that an imprisoned resident will display psychotic symptoms or report a prior diagnosis of schizophrenia. We found a similar association using the number of atypical prescriptions per capita.
Prior authorization policies were instituted to reduce spending in state Medicaid programs by steering utilization to lower-cost medications and by assisting Medicaid programs in negotiating rebates with pharmaceutical companies. This study contributes to the discussion of costs and benefits of a prior authorization policy by providing some direct evidence on the nonmedical costs of prior authorization related to increased contact with law enforcement. The analysis suggests that prior authorization and the associated reduction in atypical use may cause increased imprisonment of schizophrenic persons. The total annual cost in the United States associated with severe psychiatric disorders in jails and prisons is estimated by the Department of Justice’s Source Book of Criminal Justice Statistics11 to be $8.5 billion (based on an estimatedcost of $50,000 per ill inmate per year, with about 170,000 individuals with serious psychiatric disorders kept in jails and prisons). Adding to this expense are court costs, attorney fees, police costs, and costs of other social and medical services, suggesting that the criminal justice system is a very expensive way to deal with mental illness.Acknowledgments
The authors wish to acknowledge ApotheCom (supported by Janssen Scientific Affairs, LLC) for providing assistance with the formatting of this manuscript.
Author Affiliations: University of Southern California, Los Angeles, CA (DG, GJ, RC, DL); Janssen Scientific Affairs LLC, Titusville, NJ (JF, RD); Claremont McKenna College, Claremont, CA (EH).
Source of Funding: This research was funded by Janssen Scientific Affairs, LLC, Titusville, NJ.
Author Disclosures: During the course of this research, the following authors held positions at Precision Health Economics: DG and DL (part- ner); and GJ and RC (consultant). Precision Health Economics received funding from Janssen to complete this study. JF and RD are employees and shareholders of Janssen.
Authorship Information: Concept and design (RD, DG, GJ); acquisition of data (DG, GJ); drafting of the manuscript (EH, DL); critical revision of the manuscript for important intellectual content (DG, GJ, RD, JF, EH, RC, DL); statistical analysis (RC, DL, EH); obtaining funding (RD, JF, DG, GJ); administrative, technical, or logistic support (RC); supervision (DL, EH).
Address correspondence to: Darius Lakdawalla, PhD, Quintiles Chair in Pharmaceutical and Regulatory Innovation, University of Southern California, 3335 S Figueroa St, Los Angeles, CA 90007. E-mail: dlakda- firstname.lastname@example.org. Neuman P, Strollo MK, Guterman S, et al. Medicare prescription drug benefit progress report: findings from a 2006 national survey of seniors. Health Aff (Millwood). 2007;26(5):w630-w643.
2. Addington DE, Patten SB, McKenzie E, Addington J. Relationship between relapse and hospitalization in first-episode psychosis. Psychiatr Serv. 2013;64(8):796-799.
3. Kane JM. Improving treatment adherence in patients with schizophrenia. J Clin Psychiatry. 2011;72(9):e28.
4. Panish J, Karve S, Candrilli SD, Dirani R. Association between adherence to and persistence with atypical antipsychotics and psychiatric relapse among US Medicaid-enrolled patients with schizophrenia. J Pharm Health Serv Res. 2013;4(1):29-39.
5. Czobor P, Volavka J, Derks EM, et al; EUFEST Study Group. Insight and hostility as predictors and correlates of nonadherence in the European First Episode SchizophreniaTrial. J Clin Psychopharmacol. 2013;33(2):258-261.
6. Bourdon KH, Rae DS, Locke BZ, Narrow WE, Regier DA. Estimating the prevalence of mental disorders in U.S. adults from the Epidemiologic Catchment Area Survey. Public Health Rep. 1992;107(6):663-668.
7. Almond S, Knapp M, Francois C,Toumi M, BrughaT. Relapse in schizophrenia: costs, clinical outcomes and quality of life. Br J Psychiatry. 2004;184:346-351.
8. Law MR, Soumerai SB, Ross-Degnan D, Adams AS. A longitudinal study of medication nonadherence and hospitalization risk in schizophrenia. J Clin Psychiatry. 2008;69(1):47-53.
9.Teplin LA. Criminalizing mental disorder: the comparative arrest rate of the mentally ill. Am Psychol. 1984;39(7):794-803.
10. State of Connecticut Offic of Policy and Management. 2010 Annual Recidivism report.Hartford, CT:Criminal justice policy and planning division; 2010. Available at: http://www.ct.gov/opm/lib/opm/cjppd/cjre- search/recidivismstudy/2010_0215_recidivismstudy.pdf.
11. Maguire K, Pastore AL, eds. Sourcebook of criminal justice statistics 1996: U.S. Dept. of Justice, Bureau of Justice Statistics. Washington, DC: US6PD;1997.
12. Freedman R. Schizophrenia. N Engl J Med. 2003;349(18):1738-1749.
13. Csernansky JG, Schuchart EK. Relapse and rehospitalisation rates in patients with schizophrenia: effects of second generation antipsychotics. CNS Drugs. 2002;16(7):473-484.
14. Kishimoto T, Agarwal V, Kishi T, Leucht S, Kane JM, Correll CU. Relapse prevention in schizophrenia: a systematic review and meta-analysis of second-generation antipsychotics versus first-generation antipsychotics. Mol Psychiatry. 2013;18(1):53-66.
15. Nnadi CU, Malhotra AK. Individualizing antipsychotic drug therapy in schizophrenia: the promise of pharmacogenetics. Curr Psychiatry Rep. 2007;9(4):313-318.
16. Manschreck,TC, Boshes RA.The CATIE schizophrenia trial: results, impact, controversy. Harvard Rev Psychiatry. 2007;15(5):245-258.
17. StroupTS, Lieberman JA, McEvoy JP, et al. Results of phase 3 of the CATIE schizophrenia trial. Schizophrenia Res. 2009;107(1):1-12.
18. Basu A, Jena AB, PhilipsonTJ.The impact of comparative effectiveness research on health and health care spending. J Health Econ. 2011; 30(4):695-706.
19. Tabarrok A, Heaton P, Helland E.The measure of vice and sin: a review of the uses, limitations and implications of crime data. In: Handbook on the Economics of Crime. Northampton, MA: Edward Elgar; 2010:53-81.
20. Vogt WB, Joyce G, Xia J, Dirani R, Wan G, Goldman DP. Medicaid cost control measures aimed at second-generation antipsychotics led to less use of all antipsychotics. Health Aff (Millwood). 2011;30(12): 2346-2354.
21. King G, Zeng L. Logistic regression in rare events data. Political Analysis. 2001;9(2):137-163.