PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
J Am Geriatr Soc. Author manuscript; available in PMC 2014 May 1.
Published in final edited form as:
PMCID: PMC3656141
NIHMSID: NIHMS443905

Fracture Risk among Nursing Home Residents Initiating Antipsychotic Medications

Abstract

Objectives

to determine whether antipsychotic medication initiation is associated with subsequent fracture in nursing home residents, whether fracture rates differ between first-generation versus second-generation antipsychotic use, and whether fracture rates differ among users of haloperidol, risperidone, olanzapine, and quetiapine.

Design

time-to-event analyses were conducted in a retrospective cohort using linked Medicaid, Medicare, Minimum Data Set and Online Survey, Certification and Reporting data sets.

Setting and Participants

nursing home residents aged ≥ 65 years in CA, FL, MO, NJ and PA.

Measurements

fracture outcomes (any fracture; hip fracture) in first-versus second-generation antipsychotic users, and specifically among users of haloperidol, risperidone, olanzapine and quetiapine. Comparisons incorporated propensity scores that included patient-level variables (demographics, comorbidity, diagnoses, weight, fall history, concomitant medications, cognitive performance, physical function, aggressivebehavior) and facility-level variables (nursing home size, ownership factors, staffing levels).

Results

Among 8,262 subjects (within 4,131 pairs), 4.3% suffered any fracture during observation with 1% having a hip fracture during an average follow up period of 93 ± 71 days; range 1 to 293 days). Antipsychotic initiation was associated with any fracture (hazard ratio (HR) 1.39, p=0.004) and with hip fracture (HR 1.76, p=0.024). The highest risk was found for hip fracture when antipsychotic use was adjusted for dose(HR=2.96; p=0.008). However, no differences in time-to-fracture were found in first-versus second-generation agents or across competing individual drugs.

Conclusion

Antipsychotic initiation is associated with fracture in nursing home residents, but risk does not differ across commonly used antipsychotics.

Keywords: fracture, antipsychotic, nursing home

INTRODUCTION

Hip fractures negatively impact survival and health status.1-3 Hip fracture is four times more common in nursing home (NH) residents than in community-dwelling older adults.4 Psychotropic medications are associated with fall-related fractures in older adults.5 Along with sedating effects and orthostatic hypotension, first-generation antipsychotics (FGAP) have extrapyramidal side effects that may reduce gait stability. Second-generation antipsychotics (SGAP) have fewer extrapyramidal side effects and thus have been widely presumed to be safer for gait stability than FGAP. This distinction between FGAP and SGAP has been a central theme in the rapid transition in antipsychotic prescribing practices in the United States.6 In 2001, only seven years after the introduction of risperidone to market, SGAPs accounted for more than 90% of all antipsychotic use in NH residents.7

However, the widely-held notion that SGAPs are safer for gait stability and must therefore pose less fracture risk than FGAPs has not been shown in randomized clinical trials (RCTs). Fractures are relatively uncommon events that do not occur in sufficient numbers for evaluation during typical short-term RCTs that are designed to test drug efficacy. Published observational studies to date have reached varying conclusionsfrom different populations, study designs, available variables and AP use measures.8-13

Despite Food and Drug Administration (FDA) safety warnings and regulatory scrutiny of antipsychotic (AP) use in NHs, one in five NH residents still receive APs.14 Among residents with cognitive impairment and behavioral problems, the rate of AP usein 2010 was 39%.14 Because of the paucity of other pharmacologic alternatives, knowledge of the comparative safety across competing APs remains important for the foreseeable future.

Our goal was to use existing observational data to examine the association between AP use and fracture outcomes in NH residents, using propensity-based matching to control for patient-level and facility-level confounding factors.

METHODS

Cohort Formation

We conducted a retrospective cohort study with a new-user design in dually Medicaid- and Medicare-eligible NH residents in California, Florida, Missouri, New Jersey and Pennsylvania. Data from 1999 were selected because more recent years did not provide sufficient observations of FGAP use, owing to rapidly changing prescribing patterns. We purchased data from the Centers for Medicare and Medicaid Services (CMS) and linked Medicaid, Medicare, and Minimum Data Set (MDS) nursing home data at the individual level. We incorporated facility-level characteristics from the Online Survey, Certification and Reporting (OSCAR) database. 15

We required all subjects to have Medicare parts A and B in order to capture both inpatient and outpatient fractures. Exclusion criteria included age <65 years, NH use for <100 days, history of psychiatric hospitalization or severe mental illness, or a diagnosis of schizophrenia or bipolar mood disorder. Other pragmatic exclusions were residence in multiple states or multiple NHs, lack of a baseline MDS assessment prior to AP initiation, or lack of required data elements needed for the calculation of propensity scores or drug exposure measures. We also excluded subjects who did not have at least 60 days of observable time in the data set without AP use, in order to assure an observed ‘wash out’ period prior to AP initiation.

Exposure and Outcome Measures

National drug codes (NDC) were extracted from Medicaid pharmacy claims and converted to specific medications and doses using the Multum drug data base.16FGAPs included haloperidol, thioridazine, chlorpromazine, fluphenazine, perphenazine, thiothixene, loxapine, trifluoperazine, mesoridazine, molindone, pimozide and promazine. SGAPs included risperidone, olanzapine, and quetiapine; we excluded clozapine because of its narrow indications. AP-users who received only a single AP agent throughout observation were classified as either FGAP or SGAP users, and were further assigned into individual drug groups (risperidone, quetiapine, olanzapine, haloperidol or ‘other’). Persons who switched APs during observation were included in the AP-use versus non-use analyses but excluded from drug-drug comparisons.

The primary explanatory variable was use or non-use of any AP, but we also characterized AP exposure by regularity of use over time and dosing intensity. Use over time was described by the ‘proportion of days covered’ (PDC), a measure of medication persistence.17We allowed a seven day gap length between prescriptions and assumed that APs were continued during periods of hospitalization.18 Doses were standardized across APs of different potencies by converting average daily doses to a ‘chlorpromazine equivalent.’19

Fracture outcomes were captured from diagnosis and procedure codes within Medicare data using published methods.20-23Hip fracture was defined by a new inpatient claim for hip fracture. Non-hip fractures were deemed present according to a previously published algorithm that required a set of confirmatory imaging and physician service codes.23

Propensity Score Development and Matching

Propensity score models were developed to describe the likelihood of each person to have been an AP-user, and then used in matching as described below.24Candidate variables for propensity score modeling were based on published literature regarding patient-level and facility-level predictors of AP use and fracture in NH residents.5,7,12,22,25-28 These variables were derived from the baseline MDS record which was defined as the MDS assessment closest to but still preceding the onset of observed follow up time (i.e., prior to AP initiation among AP-users.) Demographic factors included state, gender, age, race/ethnicity and rural/urban residence. Disease burden was characterized by a claims-based version of the Charlson comorbidity index and by the total number of medications used. Dementia diagnoses and Parkinson’s disease were identified through a combination of claims, MDS markers, and tracer medications (e.g., cholinesterase inhibitor to indicate dementia). From the MDS, we identified the occurrence of a fall in the last 180 days, body mass index, urinary incontinence, and use of anxiolytic, antidepressant or hypnotic medications. Widely-used, validated MDS scales were used to characterize cognitive performance (the MDS Cognitive Performance Score),29 activities of daily living (the MDS hierarchical Activities of Daily Living Score),30 aggressive behavior (the MDS Aggressive Behavior Score)31 and mortality risk (the MDS CHESS mortality risk score which incorporates medical activity and treatments along with other domains).32Scores were converted to categorical variables by severity-based groupings. Facility-level variables included ownership factors, number of beds, hours of certified nursing assistant care and total nursing hours of care per bed per day.

To derive propensity scores, we used bivariate analyses to evaluate the balance of measured covariates between AP-users and non-users. We then used unconditional logistic regression to derive models predicting AP use as a function of these explanatory factors, with propensity scores being the resulting predicted probabilities. A propensity model with all possible two-way interaction terms was selected due to its superior performance,and the Hosmer-Lemeshow goodness-of-fit test was conducted.33

These propensity scores were then used to create 1:1 matched pairs of AP-users and similar non-users, using the greedy matching algorithm of Parsons.34 Matching was conducted within state, age group, and gender strata so the pairs were perfectly balanced on these three measures. AP-users were discarded if an appropriate non-user match could not be found. After matching, the AP-User and non-user groups were re-evaluated to assure equalization of the previous differences between groups.24

For drug-drug comparisons in AP users only, we used the same variables to calculate a new propensity score that modeled the likelihood for any individual AP user to have received a SGAP. These propensity scores were incorporated into fracture outcome analyses for purposes of covariate adjustment among AP users.

Outcome Analyses

Matched 1:1 AP-user/non-user pairs were followed in a time-to-event analysis, starting from the first day of an AP prescription (i.e., the Index Date)after the required washout period. Non-users were also required to have a similar period of observed time prior to the start date, in order to assure that both members of the pair had a similar opportunity to contribute follow-up time. Subjects were followed until a first fracture event and were right-censored if they died, were hospitalized for ≥ 30 days for reasons other than fracture, were discharged from the NH, ceased to be Medicaid or Medicare eligible, or if the AP-user in the pair stopped receiving AP medication for at least one month.

McNemar’s test was used to evaluate whether the fracture rate was greater among AP-users than among non-users. Kaplan-Meier curves were generated to visually examine time-to-fracture data. Stratified Cox proportional hazards models were used to analyze survival endpoints, using the propensity-matched pairs as the strata for adjustment for confounding.24 Hazard ratios were derived from these models. Cox proportional hazards regression models were also used to examine the impact of PDC and dose on fracture outcomes among AP users.

Fracture outcomes were also analyzed by drug class. We derived adjusted hazard ratios for fracture outcomes in SGAP users as compared to FGAP users, adjusted for the propensity to have been an SGAP user. Similarly adjusted models were used to compare fracture outcomes across individual drug groups (risperidone, olanzapine, quetiapine, haloperidol), using haloperidol as the reference group. These analyses were carried out only in subjects who stayed on the same specific AP throughout observation.

Finally, we considered the possibility of barriers to medication ascertainment in subjects who used skilled nursing facility (SNF) days during their observation period; if drugs were provided by Medicare during post-acute care, AP use would not have been ascertainable via Medicaid claims during that time. We conducted a sensitivity analysis by excluding subjects using SNF services during their period of observation, and then repeating all outcome analyses described above.

All data management and analyses were performed using SAS version 9.1.3 (2002-2003, SAS Institute Inc., Cary, NC).

This project was approved by the Institutional Review Board of The University of Kansas Medical Center. The Centers for Medicare & Medicaid Services provided data under Data Use Agreement #16798.

RESULTS

Sample Characteristics

Data files could be linked across all three data sets for 224,764 persons. After application of initial demographic and clinical inclusion/exclusion criteria, 197,514 persons remained. Of these, 130,309 persons had all required data elements needed to calculate a propensity score. The propensity model predicting AP use had a c-statistic of 0.739 with no evidence to suggest problems with goodness of fit (p=0.136). After additional exclusions due to prevalent AP use, lack of the necessary washout period prior to AP initiation, and/or lack of a baseline MDS assessment prior to AP initiation, there remained 44,898 persons with complete data. This group was comprised of 4,545 AP-users and 40,353 non-users from which we drew the matched pairs. We excluded 9% (414 out of 4,545) of AP-users for whom no match could be found. The final sample consisted of 4,131 matched 1:1 pairs of AP-users and non-users (total n=8,262).

Table 1 demonstrates differences between AP-user and non-user groups prior to matching and shows resolution of those differences after matching. The final AP-users and non-users within the 4,131 matched pairs were similar in demographic, clinical, cognitive, functional, behavioral and NH factors.

Table 1
Antipsychotic Users & Non-Users Before and After Propensity-based Matching

Antipsychotic Medication Use

As shown in Table 1, 46% of persons in the matched pairs were aged ≥85 years. Women comprised 72% of the sample, and most were cognitively and functionally impaired. Among AP users, 54% received risperidone, 15% received olanzapine, 11% received haloperidol, and 4% received quetiapine. The remaining 16% switched APs during observation or took a variety of rarely used, older FGAPs.

Fracture Outcomes in Matched Pairs

Any fracture occurred in 342 (4.3%) of subjects, with 79 (1%) having a hip fracture. Fractures were associated with AP-use. Any fracture occurred in 204 AP-users whose matched non-user did not have any fracture; in contrast, any fracture occurred in 138 non-users whose matched AP-user did not have any fracture (McNemar’s test, p<0.001). Hip fracture occurred in 52 AP-users whose matched non-user did not have a hip fracture; in contrast, hip fracture occurred in 27 non-users whose matched AP-user did not have a hip fracture (McNemar’s test, p=0.005).

Table 2 shows results of the stratified Cox regression models. AP use, compared to non-use, was associated with a hazard ratio (HR) of 1.39 (p=0.004) for any fracture, with higher HRs found when AP exposure was characterized by regularity of use over time (by PDC) or dosage (in chlorpromazine equivalents). A similar pattern was seen for hip fracture (HR=1.76; p=0.024). The highest risk was found for AP use characterized by dosage (HR=2.96; p=0.008), with this HR reflecting the increased risk for hip fracture associated with every 100 mg dose increase in chlorpromazine equivalents.

Table 2
Hazard Ratios for Any Fracture and Hip Fracture in Antipsychotic Users versus Non-Users

Fracture Outcomes in Comparator Drug Groups

As shown in Table 3, we found no difference in propensity-adjusted time-to-fracture between FGAP and SGAP users for any fracture (adjusted hazard ratio=1.016; p=0.9399) or for hip fracture (adjusted hazard ratio=0.79; p=0.529). Similarly, no differences were detected in propensity-adjusted time-to-fracture for users of haloperidol, risperidone, olanzapine or quetiapine (overall test across four drugs: p=0.825 for any fracture; p=0.529 for hip fracture). Table 3 also shows individual drug-drug comparisons using haloperidol as the reference group.

Table 3
*Adjusted Hazard Ratios for Any Fracture and Hip Fracture in Users of Different Antipsychotic Medications

Sensitivity Analysis

Finally, the pattern of these results was not altered by the exclusion of subjects who used SNF services (data not shown); this occurred in 13% of the matched pairs.

DISCUSSION

We found that risk for any fracture is ~40% higher among NH residents initiating an AP, compared to otherwise similar persons not taking an AP, when matched on a comprehensive group of confounding factors. For hip fracture specifically, the risk is ~80% higher in AP users than in non-users. Persons with ongoing regular AP use are at even greater risk, as are persons receiving higher AP doses. For each 100 mg increase in chlorpromazine equivalents, there is a ~3-fold increased risk of hip fracture. To place this in clinical context, such a dose increase equates to 2 mg of risperidone, 2 mg of haloperidol, 5 mg of olanzapine, or 75 mg of quetiapine.35

Importantly, we did not detect a difference in fracture rate between FGAP versus SGAP users or among users of the four most commonly used individual APs in propensity-adjusted analyses. These results do not support the widely held clinical notion that SGAPs are safer than FGAPs for fall/fracture risk reduction in NH residents.

Other investigators have studied associations between AP use and fracture using varied study samples, data sources and designs. Cohort studies from NH settings are most relevant. Huybrechts et al found that hip fracture occurred more often during FGAP use than in SGAP use in NH residents in British Columbia; however, functional, cognitive and behavioral data were not available for that analysis.12 Huybrechts et al also examined numerous important outcomes (hip fracture, myocardial infarction, stroke, bacterial infection, and overall mortality) associated with FGAP and SGAP use in a large cohort study of U.S. nursing homes from 2001-2005, using the same data sources incorporated here.36,37Although the point estimate suggested a trend toward a higher rate of hip fracture in FGAP users as compared to SGAP users (hazard ratio 1.29; CI 0.95-1.76), this finding did not reach statistical significance.37

Other investigators have conducted case-control studies using similar data sources. Liperoti et al used the Systematic Assessment of Geriatric Drug Use via Epidemiology (SAGE) data base to identify a 35-37% increased risk of hip fracture with both SGAP use and FGAP use.10 Jalbert et al used Medicaid and MDS data to identify an overall 26% increased risk of hip fracture with any AP use, with higher risk associated with longer periods of use.11 Among prevalent AP users, they found an association between hip fracture and SGAP use but a similar trend for FGAP use did not reach statistical significance.

Although each aforementioned paper has methodological strengths, none to date has incorporated all these elements simultaneously, as this study does : a) inclusion of cognitive, functional, and behavioral data, along with facility-level factors; b) a cohort design using propensity-matched AP users and comparable non-users; c) characterization of drug persistence over time; d) a ‘new user’ design to reduce bias inherent in prevalent user samples; e) standardization of AP dose equivalency across drugs; f) capture of fall-related non-hip fractures in addition to hip fractures; g) sufficient observations of haloperidol use to support its analysis as an individual drug group. This final point is important because haloperidol is the least sedating FGAP and the most likely candidate among FGAPs for use in older adults.

Taken together, the existing literature and the results presented here reaffirm that fracture is associated with AP use but provide no convincing evidence that fracture rates differ meaningfully across APs when detailed cognitive, functional and behavioral person-level data from the MDS are incorporated into the analysis.

LIMITATIONS

Treatment selection bias impacts all observational pharmacoepidemiologic studies. However, when assessing relatively rare events or populations that are difficult to enroll in RCTs, rigorous observational studies constitute essential evidence.38,39We addressed treatment selection bias by incorporating numerous confounders and achieving elimination of measurable differences between AP users and non-users by propensity-based matching. However, unmeasured confounding can never be completely eliminated.Although some investigators advocate the use of high-density propensity scores derived from all available data, we preferred to use literature-based guidance to select variables for inclusion. Recognizing that the quality of MDS variables for research purposes is variable, we chose only variables that were adequately supported by the literature. Differences between the results of this traditional method of propensity score development and that of the newer high-density propensity scores are modest.40

An additional limitation is that comorbidity and other adjustment variables were measured only at baseline but are not static measures. Certain clinically important factors could not be captured in these data sources. Although we lacked bone density data, osteoporosis afflicts 85% of 85-year-old women NH residents41and would have been unlikely to be differentially distributed among AP-users and non-users. We had nodirect measurement of gait and balance, but did use fall history to provide a directly relevant balance-related fracture risk measure.

Our study population was limited to dually-eligible Medicare and Medicaid beneficiaries, but such persons account for more than 50% of the NH population in the United States.14Finally, the older age of these 1999 data presents limitations. These APs may be prescribed differently at present than they were in 1999, and newer SGAPs (e.g., ziprasidone, aripiprazole) were not available during our study timeframe.

Clinical Relevance

These results contribute to changing perspectives about the relative risks of FGAPs versus SGAPs. A decade ago, it seemed a foregone conclusion that SGAPs were safer than FGAPs for older adults who required AP treatment.6Subsequently, newer data on adverse metabolic effects of SGAPs began to temper this view. A number of additional AP-associated risks, including increased mortality, cerebrovascular and cardiovascular morbidity, and pneumonia now complicate AP-related prescribing decisions.40 The FDA has issued warnings for both FGAPs and SGAPS about increased mortality and morbidity with AP use in older adults with dementia, emphasizing that all such use is off-label.42 Despite these risks, a subgroup of NH residents with dementia will require pharmacotherapy to manage distressing psychosis, in the face of limited medication alternatives.

The risk of fracture must be part of this complex prescribing decision because antipsychotic initiation is indeed associated with increased fracture risk in NH residents. However, these results suggest that fracture risk does not differ according to drug class or across the four most commonly used APs. On the other hand, fracture risk does increase with higher dose and with greater regularity of use. Prescribers should use minimal AP doses for the shortest durations possible when non-pharmacologic approaches have failed to manage dementia-related psychosis.

ACKNOWLEDGMENTS

Funding was provided by the National Institute of Mental Health under R01MH079221.

Financial Disclosure:funded by the National Institute of Mental Health under grant number R01 MH079221.

Footnotes

Conflict of Interest: The editor in chief has reviewed the conflict of interest checklist provided by the authors and has determined that the authors have no financial or any other kind of personal conflicts with this paper.

Author Contributions: Rigler: Conception and design, acquisition of data, analysis and interpretation of data, statistical analysis, preparation of the manuscript.

Shireman: Study design, analysis and interpretation of data, preparation of the manuscript.

Cook-Wiens: analysis and interpretation of the data; statistical analysis; preparation of the manuscript.

Ellerbeck: Study design, analysis and interpretation of data, preparation of the manuscript.

Whittle:study design, analysis and interpretation of the data, preparation of the manuscript.

Mehr: Conception and design, preparation of the manuscript.

Mahnken: Conception and design, statistical analysis, analysis and interpretation of data, preparation of the manuscript.

Sponsor’s Role: The Centers for Medicare & Medicaid Services provided data under Data Use Agreement #16798. Neither the funding sponsor nor CMS had any role in the study design, methods, conduct of the study, analysis or interpretation of results, or decision to publish.

Meeting Presentation: Results were presented in preliminary form as an oral research presentation on May 14, 2011 at the 2011 national annual meeting of the American Geriatrics Society, Washington, D.C. It won the “Best Paper in Epidemiology” designation at that meeting.

REFERENCES

1. Zimmerman S, Chandler JM, Hawkes W, et al. Effect of fracture on the health care use of nursing home residents. Arch Int Med. 2002;162:1502–1508. [PubMed]
2. Young Y, Fried LP, Kuo YH. Hip fractures among elderly women: Longitudinal comparison of physiological function changes and health care utilization. J Am Med Dir Assoc. 2010;11:100–105. [PMC free article] [PubMed]
3. Haentjens P, Magaziner J, Colon-Emeric CS, et al. Meta-analysis: Excess mortality after hip fracture among older women and men. Ann Int Med. 2010;152:380–390. [PMC free article] [PubMed]
4. Sugarman JR, Connell FA, Hansen A, et al. Hip fracture incidence in nursing home residents and community-dwelling older people, Washington State, 1993-1995. J Am Geriatr Soc. 2002;50:1638–1643. [PubMed]
5. Girman CJ, Chandler JM, Zimmerman SI, et al. Prediction of fracture in nursing home residents. J Am Geriatr Soc. 2002;50:1341–1347. [PubMed]
6. Consensus statement on improving the quality of mental health care in U.S. nursing homes: Management of depression and behavioral symptoms associated with dementia. J Am Geriatr Soc. 2003;51:1287–1298. [PubMed]
7. Briesacher BA, Limcangco MR, Simoni-Wastila L, et al. The quality of antipsychotic drug prescribing in nursing homes. Arch Int Med. 2005;165:1280–1285. [PubMed]
8. Kolanowski A, Fick D, Waller JL, et al. Outcomes of antipsychotic drug use in community-dwelling elders with dementia. Arch Psychiatr Nurs. 2006;20:217–225. [PubMed]
9. Mehta S, Chen H, Johnson ML, et al. Risk of falls and fractures in older adults using antipsychotic agents: A propensity-matched retrospective cohort study. Drugs Aging. 2010;27:815–829. [PubMed]
10. Liperoti R, Onder G, Lapane KL, et al. Conventional or atypical antipsychotics and the risk of femur fracture among elderly patients: Results of a case-control study. J Clin Psychiatry. 2007;68:929–934. [PubMed]
11. Jalbert JJ, Eaton CB, Miller SC, et al. Antipsychotic use and the risk of hip fracture among older adults afflicted with dementia. J Am Med Dir Assoc. 2010;11:120–127. [PubMed]
12. Huybrechts KF, Rothman KJ, Silliman RA, et al. Risk of death and hospital admission for major medical events after initiation of psychotropic medications in older adults admitted to nursing homes. Can Med Assoc J. 2011;183:E411–419. [PMC free article] [PubMed]
13. Hien LT, Cumming RG, Cameron ID, et al. Atypical antipsychotic medications and risk of falls in residents of aged care facilities. J Am Geriatr Soc. 2005;53:1290–1295. [PubMed]
14. Centers for Medicare & Medicaid Services; [Accessed July 17, 2012]. 2010. MDS 2.0 Public Quality Indicator and Resident Reports. Available at: https://www.cms.gov/Research-Statistics-Data-and-Systems/Computer-Data-and-Systems/MDSPubQIandResRep/index.html.
15. Centers for Medicare & Medicaid Services (CMS) [Accessed July 17, 2012];Institutional File: Online Survey Certification and Reporting (OSCAR) data base. Available at: http://www.cms.gov/Research-Statistics-Data-and-Systems/Files-for-Order/NonIdentifiableDataFiles/ProviderofServicesFile.html.
16. [Accessed July 17, 2012];Cerner Multum Lexicon drug database. Available at: http://www.multum.com/lexicon.htm.
17. Karve S, Cleves MA, Helm M, et al. Prospective validation of eight different adherence measures for use with administrative claims data among patients with schizophrenia. Value Health: J Int Soc Pharmacoecon Outcome Res. 2009;12:989–995. [PubMed]
18. Gardarsdottir H, Souverein PC, Egberts TC, et al. Construction of drug treatment episodes from drug-dispensing histories is influenced by the gap length. J Clin Epidemiol. 2010;63:422–427. [PubMed]
19. M Atkins AB, Bottomley C, Riccio M. Chlorpromazine equivalents: A Consensus of Opinion for Both Clinical and Research Applications. Psychiatr Bull. 1997;21:224–226.
20. Baron JA, Karagas M, Barrett J, et al. Basic epidemiology of fractures of the upper and lower limb among Americans over 65 years of age. Epidemiology. 1996;7:612–618. [PubMed]
21. Ray WA, Griffin MR, Fought RL, et al. Identification of fractures from computerized Medicare files. J Clin Epidemiol. 1992;45:703–714. [PubMed]
22. Taylor AJ, Gary LC, Arora T, et al. Clinical and demographic factors associated with fractures among older Americans. Osteopor Int. 2011;22:1263–1274. [PMC free article] [PubMed]
23. Rigler SK, Ellerbeck E, Whittle J, et al. Comparing methods to identify hip fracture in a nursing home population using Medicare claims. Osteopor Int. 2011;22:57–61. [PMC free article] [PubMed]
24. Austin PC. Assessing balance in measured baseline covariates when using many-to-one matching on the propensity-score. Pharmacoepidemiol Drug Safety. 2008;17:1218–1225. [PubMed]
25. Hughes CM, Lapane KL, Mor V. Influence of facility characteristics on use of antipsychotic medications in nursing homes. Med Care. 2000;38:1164–1173. [PubMed]
26. Chen Y, Briesacher BA, Field TS, et al. Unexplained variation across US nursing homes in antipsychotic prescribing rates. Arch Int Med. 2010;170:89–95. [PMC free article] [PubMed]
27. Kamble P, Chen H, Sherer J, et al. Antipsychotic drug use among elderly nursing home residents in the United States. Am J Geriatr Pharmacother. 2008;6:187–197. [PubMed]
28. Bolton JM, Metge C, Lix L, et al. Fracture risk from psychotropic medications: A population-based analysis. J Clin Psychopharm. 2008;28:384–391. [PubMed]
29. Morris JN, Fries BE, Mehr DR, et al. MDS Cognitive Performance Scale. J Gerontol. 1994;49:M174–182. [PubMed]
30. Morris JN, Fries BE, Morris SA. Scaling ADLs within the MDS. J Gerontol Series A: Biol Med Sci. 1999;54:M546–553. [PubMed]
31. Perlman CM, Hirdes JP. The aggressive behavior scale: A new scale to measure aggression based on the minimum data set. J Am Geriatr Soc. 2008;56:2298–2303. [PubMed]
32. Hirdes JP, Frijters DH, Teare GF. The MDS-CHESS scale: A new measure to predict mortality in institutionalized older people. J Am Geriatr Soc. 2003;51:96–100. [PubMed]
33. Austin PC. Goodness-of-fit diagnostics for the propensity score model when estimating treatment effects using covariate adjustment with the propensity score. Pharmacoepidemiol Drug Safety. 2008;17:1202–1217. [PubMed]
34. Parsons L. Reducing bias in a propensity score matched-pair sample using greedy matching techniques. Proceedings of the Twenty-sixth Annual SAS Users Group International Conference.2001.
35. Woods SW. Chlorpromazine equivalent doses for the newer atypical antipsychotics. J Clin Psych. 2003;64:663–667. [PubMed]
36. Huybrechts KF, Gerhard T, Crystal S, et al. Differential risk of death in older residents in nursing homes prescribed specific antipsychotic drugs: Population based cohort study. BMJ. 2012;344:e977. [PMC free article] [PubMed]
37. Huybrechts KF, Schneeweiss S, Gerhard T, et al. Comparative safety of antipsychotic medications in nursing home residents. J Am Geriatr Soc. 2012;60:420–429. [PMC free article] [PubMed]
38. Concato J, Shah N, Horwitz RI. Randomized, controlled trials, observational studies, and the hierarchy of research designs. N Engl J Med. 2000;342:1887–1892. [PMC free article] [PubMed]
39. Schneeweiss S, Patrick AR, Sturmer T, et al. Increasing levels of restriction in pharmacoepidemiologic database studies of elderly and comparison with randomized trial results. Med Care. 2007;45(10 Supl 2):S131–142. [PMC free article] [PubMed]
40. Huybrechts KF, Brookhart MA, Rothman KJ, et al. Comparison of different approaches to confounding adjustment in a study on the association of antipsychotic medication with mortality in older nursing home patients. Am J Epidemiol. 2011;174:1089–1099. [PMC free article] [PubMed]
41. Zimmerman SI, Girman CJ, Buie VC, et al. The prevalence of osteoporosis in nursing home residents. Osteopor Int. 1999;9:151–157. [PubMed]
42. [Accessed July 17, 2012];Antipsychotics, Conventional and Atypical. 2008 Jun 16; Available at; http://www.fda.gov/Safety/MedWatch/SafetyInformation/SafetyAlertsforHumanMedical Products/ucm110212.htm.