Medical Costs After Initiation of Drug Treatment for Overactive Bladder: Effects of Selection Bias on Cost Estimates

Published on: 
, , , ,
Supplements and Featured Publications, New Perspectives on Overactive Bladder: Quality of Life Impact, Medication Persistency, and Treatmen, Volume 11, Issue 4 Suppl

Objectives: To compare posttreatment medicalcosts for patients with overactive bladder (OAB)initiating treatment with oxybutynin chloride immediaterelease (oxybutynin IR), oxybutynin chlorideextended release (oxybutynin ER), or tolterodineextended-release tartrate capsules (tolterodine ER).

International Statistical

Classification of Diseases, Ninth Revision, Clinical

Modification (ICD-9-CM)

Methods: Data were drawn from administrativeclaims of enrollees aged 18 years and older of alarge US health plan. OAB patients were identified ifat least 1 claim with an code for OAB appeared inmedical claims from January 1, 2001, to December31, 2002. The index prescription was assigned as thefirst filled prescription of oxybutynin IR (n = 3052),oxybutynin ER (n = 4503), or tolterodine ER (n =7027) during the subject identification period.Medical costs over the year after initiation were calculatedas a function of the health plan and memberliability. Independent variables were treatmentcohort, sex, age group, geographic region, baselinecosts, specific OAB diagnosis codes, and comorbidillnesses. To compare medical costs across treatmentcohorts, multivariate regressions correcting forpotential selection bias were used.



Results: Multivariate analysis results revealedthat costs for patients taking oxybutynin IR were48% higher than costs for patients taking tolterodineER (= .026), and costs for patients taking oxybutyninER were 191% higher than costs for patientstaking tolterodine ER (<.0001). Adjusted medicalcosts were $7486 for patients taking oxybutynin IRand $14 766 for patients taking oxybutynin ERcompared with $5074 for patients taking tolterodineER.

Conclusion: Differences in medical costs thatremained after adjusting for patient characteristicssuggest that treatment with tolterodine ER may beassociated with lower medical care utilization afterinitiation of therapy for OAB.

(Am J Manag Care. 2005;11:S130-S139)

Despite recognition of the prevalence ofoveractive bladder (OAB) and its costlyconsequences in terms of patientoutcomes, there is a dearth of evidence on thecost effectiveness of the most common pharmacologicaltherapies for this condition.The overall prevalence of OAB in theUnited States has been estimated at 16.9%for women (9.3% with frequency, urgency,and urgency incontinence) and 16% for men(2.6% with frequency, urgency, and urgencyincontinence).1 These figures translate toapproximately 33.3 million adults in theUnited States with OAB and 12.2 millionadults with urgency incontinence.

The cost of OAB in the United States isestimated to be greater than $12 billionannually, a burden that is comparable withother chronic conditions, such as osteoporosis($13.8 billion).2,3 In contrast to manylong-term diseases, a large portion of thecost of OAB ($3.9 billion) is related to indirectcosts beyond the costs of the conditionitself.2,4 This is caused, in part, by the higherrisk faced by patients with OAB for experiencingrelated comorbidities.4-6 A recentanalysis of the prevalence of falls and fractures,urinary tract infections, skin infections,vulvovaginitis, and depression among23 112 OAB patients and matched controls,found that patients with OAB were significantlymore likely than controls to haveeach of the studied comorbidities.7Compared with the general population,patients with OAB may also incur greatercosts in managing these conditions.7

Although management of OAB may involvenonpharmacologic methods, antimuscarinicagents are the primary treatment for OAB.8-12These agents appear to improve both clinicaloutcomes and patients' quality of life,13-21 andthere is some indication that use of pharmacotherapyas initial treatment is associatedwith decreased utilization of healthcare resources.22 Further, better treatment of OAB isexpected to decrease the incidence of healthcareresource utilization associated withOAB-related comorbidities. Unfortunately,treatment rates are low because of a number offactors, including a reluctance to discuss OABsymptoms with a physician23; possible physicianundertreatment of OAB23; and low rates ofadherence and persistence with OAB therapy.24-27 The extent to which increasing treatmentrates may reduce healthcare costs forpatients with OAB depends on the impact ofthe specific therapies used on costs.

Few studies have directly examined theimpact of specific pharmaceutical treatmentsfor OAB on expenditures for medicalcare in a "real-world" population. Specifically,with the introduction of extendedreleaseformulations of both oxybutynin andtolterodine, there are little data quantifyingthe extent to which these newer formulationsmay reduce medical costs for patientswith OAB, or comparing the relative effectsof the extended-release formulations oncosts. Hall and colleagues found that althoughaverage per-patient-per-month totalhealthcare costs increased after the firstindication of OAB, patients filling prescriptionsfor tolterodine immediate release(tolterodine IR) or oxybutynin immediaterelease (oxybutynin IR) did not have a statisticallysignificant difference comparedwith untreated patients in the percentagechange in total healthcare costs.25 However,a lengthier follow-up time than was availablein that study may be necessary to fully evaluatethe impact of OAB therapies on costs.

This study assessed medical cost outcomesfor patients with OAB who initiatedtreatment with tolterodine tartrate capsulesextended release (tolterodine ER) comparedwith patients initiating treatment with oxybutyninchloride extended release (oxybutyninER) or oxybutynin IR.


Study Design, Data Source, and Study


This retrospective study usedeligibility, medical, and pharmacy claims ofa large national US health plan affiliated withIngenix. The research database includes discounted,fee-for-service, independent, practice-associated model plans covering allregions of the United States. During thestudy period, approximately 5.7 millionpatients older than the age of 18 years hadclaims in the database.


Classification of Diseases, Ninth

Revision, Clinical Modification (ICD-9-CM)

Patients were included in the study if theyhad commercial or Medicare insurance witha prescription drug benefit. For each subject,an index date was identified as the firstfilled prescription for tolterodine ER, oxybutyninER, or oxybutynin IR during January1, 2001, to December 31, 2002. The studyperiod consisted of a 6-month baseline periodbefore and a 12-month follow-up periodafter the index prescription. Subjects whofilled prescriptions for these drugs duringthe 6-month baseline period were excluded,as were subjects who filled prescriptions formore than 1 of the 3 drugs on their indexdate. In addition, subjects with diagnosis codes for Alzheimer's disease,spinal cord injuries, or spina bifida wereexcluded (see Appendix). Subjects wererequired to be continuously covered by thehealth plan for 6 months before and 12months after their index date. Claims withOAB-related diagnosis codes were identifiedduring the subject identification period(Table 1).

Statistical Analyses.


Drug cohorts werecompared on medical costs over 12 monthsof follow-up. Medical costs were calculated asa function of health plan and member liabilityfor nonpharmacy costs, including hospital,emergency room, physician, outpatient, laboratory,and other nonpharmacy costs.Differences between cohorts in proportionswere tested using chi-square tests. Kruskall-Wallis chi-squares were used to test for differencesin distribution of costs, andBonferonni-adjusted -tests were used tocompare pairs of drug cohorts. Multivariateanalyses assessed medical costs adjusted fordrug cohort and covariates. Other covariatesincluded in multivariate models were sex,age group, geographic region, logged totalhealthcare costs in the 6-month baselineperiod, OAB-related diagnosis code, anddiagnoses of comorbid conditions. Comorbidillnesses were identified during the baselineperiod using the Washington method of clusteringillnesses.28 Specific interactions weretested to examine whether the effect of treatmentcohorts on medical costs differed bysex, age group, baseline total healthcarecosts, or specific OAB-related diagnosis code.

A major concern in estimating the costmodels was the potential for selection biasas a result of unobserved differencesbetween the treatment cohorts (ie, whetheran unobserved factor was correlated withboth the selection of a specific treatmentand with medical cost in the follow-up period).To examine this issue, we estimated thecost equation using several methods.Ordinary least squares (OLS) regression didnot include an adjustment for any potentialselection bias. Treatment effects regressionsestimated a pair of equations, with one equationto predict the selection of treatmentand another equation to estimate cost. Thistechnique estimates the correlation betweenthe equation predicting treatment and theequation predicting cost, giving an estimateof the direction and size of the bias producedby the OLS estimation. A positive correlationindicates OLS estimates are biasedupwards and a negative correlation indicatesOLS estimates are biased downward. Chowtests assessed whether the coefficients forother covariates included in the cost modeldiffered between the treatment cohorts.Treatment regression models were estimatedseparately for the comparisons of oxybutyninIR with tolterodine ER, and ofoxybutynin ER with tolterodine ER.

Log-transformed medical costs were usedas the dependent variable in all multivariateestimations to account for the non-normaldistribution of the cost data. To interpretthe coefficients in dollar terms (instead oflog dollars), coefficients and predictedmeans were obtained by retransformation.Kennedy's correction was used to generateretransformed coefficients that representthe magnitude of the effect of the independentvariables on cost in real dollar values.29After retransformation, the coefficients representthe percent change in costs associatedwith the treatment cohort compared withthe reference treatment cohort. Robust estimatesof standard errors (SEs) using White'scorrection were obtained for models violatingthe assumption of homoscedasticity inthe error term.30,31

We estimated the dollar value of theimpact of treatment drug on medical cost bymultiplying the Kennedy-corrected exponentiatedcoefficients for the oxybutynin IRand oxybutynin ER cohorts by the meancost for the tolterodine ER cohort (referencegroup). To calculate 95% confidence intervals(CIs) for the adjusted means, the upperand lower limits for Kennedy-corrected logscalecoefficients were exponentiated.


Descriptive Statistics.




A total of 14 582subjects met all inclusion criteria. Of thosesubjects, 3052 initiated oxybutynin IR, 4503initiated oxybutynin ER, and 7027 initiatedtolterodine ER. Cohorts differed in baselinecharacteristics. The tolterodine ER cohorthad the lowest proportion of men comparedwith the oxybutynin IR and oxybutynin ERcohorts (17% vs 32% and 23% respectively;<.0001). Cohorts also differed slightly ingeographic distribution. Mean age for oxybutyninIR was 55 years (standard deviation[SD] = 16.02), similar to the oxybutynin ERcohort (55 years, SD = 14.69) and slightlyolder than the tolterodine ER (54 years, SD= 14.28) cohort (<.05). Cohorts also differedon the proportion of patients with specificOAB-related codes. Meantotal costs during the 6-month baseline periodwere highest for the oxybutynin IRcohort ($4956), followed by the oxybutyninER cohort ($4146) and the tolterodine ERcohort ($3349) (Table 2).

Total unadjusted medical costs over thefollow-up period differed between cohorts.The majority of subjects in each cohortincurred some level of medical costs overthe 12 months after initiation of therapy.Among the oxybutynin IR cohort, 3% of subjectsdid not incur medical costs comparedwith 2% in the oxybutynin ER and tolterodineER cohorts, a difference which was statisticallysignificant. With a mean of $7083,subjects in the oxybutynin IR cohort had thehighest unadjusted medical costs during follow-up, followed by $5980 for the oxybutyninER cohort and $5074 for thetolterodine ER cohort. The $2010 differencein mean cost (rounded to the nearest dollar)between the oxybutynin IR and tolterodineER cohorts was statistically significant (95%CI, $924-$3096). However, differences inunadjusted mean cost were not statisticallysignificant for comparisons of the oxybutyninIR and oxybutynin ER cohorts ($1103;95% CI, $-72-$2278) and the oxybutynin ERand tolterodine ER cohorts ($906; 95% CI,$-50-$1863) (Table 3).

Multivariate Results.



Tables present statisticallysignificant covariates only; althoughadjusted for the analyses, coefficients forcomorbid illnesses are not presented becauseof the number of these covariates. The treatmentregression estimation indicated thatoxybutynin IR was associated with 48% highermedical costs during the follow-up periodcompared with tolterodine ER (Table 4). Thestatistically significant negative correlationbetween the treatment selection and costequations for this model (rho = -.1340, =.0405) suggested that OLS estimates maybe biased downward (ie, the OLS coefficientfor oxybutynin IR was too small), andthe selectivity-corrected results were likelybetter estimates. Other factors that wereassociated with medical cost (<.05)included age group, baseline total costs, specificOAB-related diagnosis code, and specificcomorbid illnesses. For each $1000increase in baseline total costs, subjects' follow-up medical costs were approximately 2%higher. Specific OAB-related diagnosis codeswere associated with varying levels of highercosts compared with subjects without thosespecific diagnosis codes. The largest effectwas for subjects with "Urethral syndromeNOS [not otherwise specified]," who had65% higher costs compared with subjectswithout that code.




Results adjusted for covariates and forselection suggest that the oxybutynin ERcohort had medical costs over the follow-upperiod that were 191% higher than for thetolterodine ER cohort (Table 5). The negativecorrelation (rho = -0.3843, = .0003)between the treatment selection equationand the cost equation suggested that theOLS estimated coefficient for oxybutyninER was biased downward and underestimatedfollow-up medical costs for the oxybutyninER. Other statistically significantcoefficients included age group, baselinetotal healthcare costs, specific OAB-relateddiagnosis code, and comorbid illnesses.Baseline total healthcare costs were associatedwith an increase of follow-up medicalcosts of approximately 2% per $1000 of baselinecosts. Subgroups with specific OAB-related diagnosis codes had higherfollow-up costs compared with subjectswithout those codes, with the largest effectfor the diagnosis of "Urethral syndromeNOS."

Mean medical costs after adjusting forcovariates and selection effects were significantlyhigher for the oxybutynin IR and oxybutyninER cohorts compared with thetolterodine ER cohort (Table 6). Resultsindicated that subjects initiating oxybutyninIR had an adjusted medical cost of $7486(95% CI, $5242-$10 690); this cost was$2412 higher than costs for subjects initiatingtolterodine ER. Subjects initiating oxybutyninER had an adjusted mean medicalcost of $14 766 (95% CI, $8501-$25 649),which is $9692 higher than the medicalcosts for subjects initiating tolterodine ER.Chow tests and interaction terms testedbetween treatment cohort and covariateswere not statistically significant in eitherestimation.


Subjects initiating oxybutynin IR, oxybutyninER, and tolterodine ER differed intheir adjusted medical costs over a 1-yearfollow-up period. Results of these analysesare particularly intriguing, because thepresence of a selection effect suggests thatanalyses of costs that fail to account for thiseffect may underestimate the extent towhich tolterodine ER may be associatedwith reduced costs after treatment initiation.The size of the treatment effects weresubstantial, with adjusted mean medicalcosts that were $2412 higher for the oxybutyninIR cohort and $9692 higher for theoxybutynin ER cohort compared with thetolterodine ER cohort.

This study addressed an important issuecommon to observational studies. Becausedata were collected retrospectively, theremay be some concern that nonrandomassignment to treatment may result in comparisonsof unbalanced populations withunobserved characteristics that may beassociated with higher costs over the followupperiod.32 This study used econometrictechniques to identify the bias in the uncorrectedestimates and to generate estimatesof the association between treatment andcost that account for this bias resulting fromunobserved factors.





One argument against examining medicalcosts alone is that the differences in medicalcost between the treatment cohorts mayhave been offset by higher prescription drugcosts for the tolterodine ER cohort. Toexamine this, models were re-estimated ontotal costs that included both pharmacy andmedical costs over the follow-up period.Results of these analyses were mixed. Forthe comparison of oxybutynin IR withtolterodine ER, the correlation between thetreatment selection and the cost equationwas no longer statistically significant (Waldchi-square = 2.52, = .112), suggesting thatthe OLS estimate of the association betweenoxybutynin IR and total cost was not biasedby treatment selection. The coefficient fromthe OLS estimation (beta = -.0803259,robust SE = 0.0246, T = -3.26, <.001)indicated that oxybutynin IR was associatedwith a 7.75% decrease in total costs comparedwith the tolterodine ER cohort (95%CI, -12.09%-3.19%). These results suggestthat although oxybutynin IR may be associatedwith higher medical costs, these may beoffset by lower pharmacy costs for thisgroup. The opposite result was found incomparing total costs for oxybutynin ERwith tolterodine ER. In the model estimatedto compare total healthcare costs for theoxybutynin ER cohort with the tolterodineER cohort, the correlation between treatmentselection equation and total costequation was statistically significant (rho =-0.664, Wald chi-square = 261.55, <.0001),indicating that there was a substantial selectionbias. The coefficient of 1.2596 (robustSE = 0.069, T = 18.23, <.0001) from thisestimation suggested that after adjusting forcovariates, subjects initiating oxybutyninER had total healthcare costs in the yearafter treatment initiation that were 251.58%higher compared with total costs for subjectsinitiating tolterodine ER (95% CI, 207.05%-302.57%).

The selection of specific therapies forOAB may represent a significant driver ofmedical costs in treating patients with OAB.For the 7027 subjects included in this studywho had initiated tolterodine ER, the 48%higher cost associated with oxybutynin IRsuggests that if this group had initiated oxybutyninIR in lieu of tolterodine ER, theywould have accrued $16 949 124 in extramedical costs over the year after treatmentinitiation. If subjects had initiated oxybutyninER in lieu of tolterodine ER, resultssuggest that an extra $68 105 684 in medicalcosts would have accrued for this group. Atthe low end, the 95% lower confidence limitof a 3.31% increase in medical costs attributableto oxybutynin IR translates to a totalof $1 180 536 in extra medical costs if the7027 tolterodine ER initiators had initiatedoxybutynin IR instead.


These findings should beinterpreted in light of the limitations of thedata and study design. The pattern of theassociation of age group with costs suggeststhat these data may underestimate costs forthe Medicare-eligible population, particularlybecause there was little difference in costamong subjects older than the age of 65years. Although these data may capture thecosts paid by patients and by the Ingenix-affiliatedhealth plan as the secondary payer,the data may not entirely capture costs thatare paid by Medicare. However, this underestimationis not expected to differ betweenthe different drug cohorts, and it is unlikelyto have affected the estimates of theassociation between drug cohort and cost.

Claims data may more accurately capturepatients' interactions with the healthcaresystem rather than patients' actual medicalhistory, and are subject to possible codingerrors and coding for the purpose of rulingout rather than actual disease. With OAB inparticular, the difference between OAB ratesseen in administrative claims data comparedwith the rates in the literature based on surveysmay, in part, be a result of underreportingof OAB. The analysis used anintent-to-treat approach, with patientsassigned to cohorts based on their first prescriptionof the period, without adjusting fordiscontinuation of medications. In addition,patients may have received drugs withoutthe presence of a prescription claim throughsamples provided by their physicians. Bothdiscontinuation and presence of sampleswould tend to reduce the estimated differencein effect of treatment on costs towardszero if patients had low adherence to theirprescribed medication, or if patients in onetreatment cohort were also taking the othertreatment. Medical claims also will not captureprovider-prescribed therapies, such asKegel exercises, habit training, or biofeedback,that are not covered by the healthplan. Since clinical efficacy measures werenot tested in this analysis, results do notimply superiority in clinical measures oroutcomes for any of the treatments.

The results are primarily applicable to thetreatment of OAB in managed care settingssimilar to the data source. However, as theplans included were discounted fee-for-serviceplans and not capitated or gatekeepermodels, the results may be more broadlyapplicable to populations covered undersimilar plans. In addition, the wide geographicdistribution provides for generalizationto managed care populations on anational level for the United States.

Despite these limitations, this examinationof outcomes associated with medicationuse in a "real-world" setting, away from thehighly controlled environment of the clinicaltrial, establishes a benchmark for the impactthat OAB therapies may have on medicalcosts in a large, insured population.


This study found that patients who initiatedeither oxybutynin IR or oxybutynin ERhad higher medical costs over 1 year of follow-up compared with patients initiatingtreatment with tolterodine ER. The extent towhich the healthcare system can reducecosts and improve outcomes for patientswith OAB may depend in part on the selectionof specific treatments as well as increasingoverall diagnosis and treatment rates.

Possible reasons behind the cost differencesbetween the therapies were notexplored in this study and remain to beinvestigated. However, there is some suggestionin the literature of lower rates of illnesses,such as fractures and urinary tractinfections for treated compared to untreatedOAB patients.6 The assessment of whether asimilar relationship may partially explainthe cost differential between therapies presentsan interesting subject for futureresearch.

World J Urol.

1. Stewart WF, Van Rooyen JB, Cundiff GW, et al.Prevalence and burden of overactive bladder in theUnited States. 2003;20:327-336.


2. Hu TW, Wagner TH, Bentkover JD, LeBlanc K, ZhouSZ, Hunt T. Estimated economic costs of overactivebladder in the United States. 2003;61:1123-1128.


3. Hu TW, Wagner TH, Bentkover JD, et al. Costs ofurinary incontinence and overactive bladder in theUnited States: a comparative study. 2004;63:461-465.

Drug Benefit Trends.

4. Zhou Z, Jensen G. Insurance claim costs for overactivebladder disorder. 2001;13:45-48.

Am J Manag


5. Wagner TH, Hu TW, Bentkover J, et al. Health-relatedconsequences of overactive bladder. 2002;8(19 suppl):S598-S607.

Am J Manag


6. Brown JS, McGhan WF, Chokroverty S. Comorbiditiesassociated with overactive bladder. 2000;6(11 suppl):574-579.


7. Darkow T, Fontes CL, Williamson TE. Costs associatedwith the management of overactive bladder and relatedcomorbidities. 2005;25:511-519.


8. Holroyd-Leduc JM, Straus SE. Management of urinaryincontinence in women: clinical applications. 2004;291:996-999.


9. Holroyd-Leduc JM, Straus SE. Management of urinaryincontinence in women: scientific review. 2004;291:986-995.

Mayo Clin Proc.

10. Elliott DS, Lightner DJ, Blute ML. Medical managementof overactive bladder. 2001;76:353-355.


11. Assessment and treatment of urinary incontinence.Scientific Committee of the First International Consultationon Incontinence. 2000;355:2153-2158.

J Clin Pharm Ther.

12. Sarkar PK, Ritch AE. Management of urinary incontinence.2000;25:251-263.

World J


13. Homma Y, Kawabe K. Health-related quality of lifeof Japanese patients with overactive bladder treated withextended-release tolterodine or immediate-release oxybutynin:a randomized, placebo-controlled trial. 2004;22:251-256.

Lancet Neurol.

14. Andersson KE. Antimuscarinics for the treatment ofoveractive bladder. 2004;3:46-53.


15. Herbison P, Hay-Smith J, Ellis G, Moore K.Effectiveness of anticholinergic drugs compared withplacebo in the treatment of overactive bladder: systematicreview. 2003;326:841-844.

Mayo Clin Proc.

16. Diokno AC, Appell RA, Sand PK, et al. Prospective,randomized, double-blind study of the efficacy andtolerability of the extended-release formulations ofoxybutynin and tolterodine for overactive bladder:results of the OPERA trial. 2003;78:687-695.

J Am Geriatr Soc.

17. Zinner NR, Mattiasson A, Stanton SL. Efficacy, safety,and tolerability of extended-release once-daily tolterodinetreatment for overactive bladder in older versusyounger patients. 2002;50:799-807.

Curr Med Res Opin.

18. Sussman D, Garely A. Treatment of overactive bladderwith once-daily extended-release tolterodine or oxybutynin:the Antimuscarinic Clinical Effectiveness Trial(ACET). 2002;18:177-184.

Mayo Clin Proc.

19. Appell RA, Sand P, Dmochowski R, et al.Prospective randomized controlled trial of extendedreleaseoxybutynin chloride and tolterodine tartrate inthe treatment of overactive bladder: results of theOBJECT Study. 2001;76:358-363.

Int Urogynecol J Pelvic Floor


20. Drutz HP, Appell RA, Gleason D, Klimberg I,Radomski S. Clinical efficacy and safety of tolterodinecompared to oxybutynin and placebo in patients withoveractive bladder. 1999;10:283-289.


21. Appell RA. Clinical efficacy and safety of tolterodinein the treatment of overactive bladder: a pooled analysis.1997;50:90-96.

Clin Ther.

22. Boone TB, Kusek JW, Nyberg LM, et al. Treatmentpatterns and associated symptom improvementduring six months of care for overactive bladder: aprospective, observational study. 2002;24:397-408.

y. BJU Int.

23. Milsom I, Abrams P, Cardozo L, et al. How widespreadare the symptoms of an overactive bladder andhow are they managed? A population-based prevalencestud2001;87:760-766.

J Manag Care Pharm.

24. Noe L, Becker R, Williamson T, Chen D. A pharmacoeconomicmodel comparing two long-acting treatmentsfor overactive bladder. 2002;8:343-352.

Manag Care Interface.

25. Hall JA, Nelson MA, Meyer JW, Williamson T,Wagner S. Costs and resources associated with thetreatment of overactive bladder using retrospectivemedical care claims data. 2001;14:69-75.


26. Lawrence M, Guay DR, Benson SR, Anderson MJ.Immediate-release oxybutynin versus tolterodine indetrusor overactivity: a population analysis. 2000;20:470-475.

Clin Ther.

27. Malone DC, Okano GJ. Treatment of urge incontinencein Veterans Affairs medical centers. 1999;21:867-877.



28. Schneeweiss R, Rosenblatt RA, Cherkin DC,Kirkwood CR, Hart G. Diagnosis clusters: a new tool foranalyzing the content of ambulatory medical care. 1983;21:105-122.


Econ Rev.

29. Kennedy PE. Estimation with correctly interpreteddummy variables in semilogarithmic equations. 1981;71:801.


30. White H. A heteroskedasticity-consistent covariancematrix estimator and a direct test for heteroskedasticity.1980;48:817-830.


31. White H. Maximum likelihood estimation of misspecifiedmodels. 1982;50:1-25.

Ann Intern Med.

32. Laupacis A, Mamdani M. Observational studies oftreatment effectiveness: some cautions. 2004;140:923-924.


33. Nitz N, Jumadilova Z, Meyer J, Bavendam T.Implications of overactive bladder codechanges and use of codes in a managed care population.Presented at: International Society of Pharmacoeconomicsand Outcomes Research; October 24-26,2004; Hamburg, Germany.