Full Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=ierp20

Expert Review of Pharmacoeconomics & OutcomesResearch

ISSN: 1473-7167 (Print) 1744-8379 (Online) Journal homepage: https://www.tandfonline.com/loi/ierp20

Direct cost of Parkinson’s disease in a healthsystem with high judicialization: evidence fromColombia

Sergio I. Prada, Ana Melissa Pérez, Jaime Valderrama–Chaparro, María IsabelMolina–Echeverry, Jorge Luis Orozco & Yuri Takeuchi

To cite this article: Sergio I. Prada, Ana Melissa Pérez, Jaime Valderrama–Chaparro, MaríaIsabel Molina–Echeverry, Jorge Luis Orozco & Yuri Takeuchi (2019): Direct cost of Parkinson’sdisease in a health system with high judicialization: evidence from Colombia, Expert Review ofPharmacoeconomics & Outcomes Research, DOI: 10.1080/14737167.2020.1681266

To link to this article: https://doi.org/10.1080/14737167.2020.1681266

View supplementary material

Published online: 18 Oct 2019.

Submit your article to this journal

Article views: 3

View related articles

View Crossmark data

ORIGINAL RESEARCH

Direct cost of Parkinson’s disease in a health system with high judicialization:evidence from ColombiaSergio I. Prada a,b, Ana Melissa Pérezb, Jaime Valderrama–Chaparroa, María Isabel Molina–Echeverryc,Jorge Luis Orozcod and Yuri Takeuchic,d

aCentro de Investigaciones Clínicas (CIC), Fundación Valle del Lili, Cali, Colombia; bCentro PROESA, Universidad Icesi, Cali, Colombia; cFacultad deCiencias de la Salud, Universidad Icesi, Cali, Colombia; dDepartamento de Neurología, Fundación Valle del Lili, Cali, Colombia

ABSTRACTObjective: To estimate all-claims-all-conditions expenditures paid for by health plans for patientssuffering from Parkinson´s disease (PD).Methods: Using administrative claims data from two health maintenance organizations for 2014 and 2015in Colombia, we identified 2,917 patients with PDby applying an algorithm that uses International StatisticalClassification of Diseases and Related Health Problems and Anatomical Therapeutic Chemical ClassificationSystem codes. Descriptive statistics were applied to compute unadjusted all-cause median costs.A generalized linear model was used to estimate adjusted and attributable direct costs of advanced PD.Results: Approximately 30% of the all-cause direct costs were associated with technologies not included inuniversal health coverage benefit packages. In 2015, the annual median interquartile range per patient all-cause direct costs to insurers was USD1,576 (605–3,617). About 16% of patients had advanced PD.Regression analysis estimated that additional costs attributable to advanced PD was USD3,416 (p = 0.000).Multimorbidity was highly prevalent, and 96% of PD patients had at least one other chronic condition.Conclusions: In the context of high judicialization, patients suffering from PD must increasingly use thejudicial system to access treatment. To promote more equitable and efficient access benefit packages,developing countries must consider more thoroughly the needs of these patients.

ARTICLE HISTORYReceived 29 January 2019Accepted 14 October 2019

KEYWORDSParkinson’s disease; cost ofillness; cost analysis;judicialization; healtheconomics

1. Introduction

Parkinson’s disease (PD) is the second most prevalent neuro-degenerative disorder in the world [1]. Studies have concludedthat PD patient expenditures are higher compared to similarpatients without the disease [2]. Inpatient care, long-term care,and drugs have been identified as major drivers of costs inEuropean countries [3]. Additionally, PD patients have higherhospital admission rates than the general population [4].

In Europe, cost-of-illness studies have reported wide varia-tions from EUR2,620 to EUR9,820 per person per year [3,5–7]including direct and indirect costs. However, it is not possible tocompare these studies without a proper meta-analysis. Suchvariations may be explained by differences in confounders,such as what is covered, the perspective (i.e., societal, hospital,family, patient), economic conditions, frequency of use, and pricedifferences.

Colombia´s health system is unique in Latin America. Colombiahas a central public fund that collects money from the workingpopulation via mandatory contributions and receives money fromgeneral taxes to cover the poor. Citizens are free to choose healthplans, which, in Colombia, are akin to health maintenance organi-zations (HMO). Each HMO gets an age-sex-region adjusted per-capita premium from the central fund to cover benefits that aregovernment defined and regulated. Private and public providerscompete for inclusion in HMO networks. In addition to benefits

financed by the central fund, patients can access drugs and pro-cedures not included in the benefits package. This is done via thejudiciary system. Patients can sue their HMO, claiming their con-stitutional right to health. It is worth noting that there is no cost fora patient to sue the government. When the judge rules in favor ofthe patient, the HMOhas to provide those services required by thepatient within a given time limit. Additionally, the patient isallowed to request reimbursement from a different fund alsomanaged by the government. To date, there has been no quanti-tative study in Colombia examining this type of judicialization bycondition, because the data is not public. Therefore, it is notpossible to knowwhether this is more common for PD treatmentsthan other treatments. Our study is the first to shed some light onquantifying expenditures for PD related to judicialization. Yet,there is evidence that, owing to a government decision to setmaximum reimbursement prices on drugs, quantities of drugsrequested for reimbursement have also been growing at higherrates [8].

The purpose of this paper is to quantify HMO expensesassociated with patients with a PD diagnoses using adminis-trative data, separating expenses covered under the universalhealth packages and those provided on an individual basis viathe judiciary system. Additionally, using the available data, weestimate costs attributable to advanced PD in the presence ofother comorbidities.

CONTACT Sergio I. Prada sergio.prada@fvl.org.co Centro de Investigaciones Clínicas (CIC), Fundación Valle del Lili, Cra.98 #18-49, Cali, Colombia

Supplemental data for this article can be accessed here

EXPERT REVIEW OF PHARMACOECONOMICS & OUTCOMES RESEARCHhttps://doi.org/10.1080/14737167.2020.1681266

© 2019 Informa UK Limited, trading as Taylor & Francis Group

2. Methods

This is a descriptive study using administrative data from2014 and 2015. These 2 years were selected, because claimsand enrollment files were completed and were available forthe study. De-identified claims from two HMOs in Colombiawere obtained under an academic agreement withUniversidad Icesi. In total, both HMOs covered 4.3 millionpeople around the country. Both HMOs in our study insuredpeople using the mandatory contributive scheme, coveringworkers and their family members. 17% of the populationwas mandatorily enrolled in the scheme as non-poor andaccounted for 49% of people enrolled in the southwestregion of Colombia. They were the third largest when mea-sured by the gross domestic product share.

2.1. Information provided by the HMOs

The HMOs provided three datasets for this study. First,they provided claims data on services and drugs renderedin all settings covered (e.g., emergency care, inpatient, andoutpatient). Claims data included the following fields:patient´s identification (ID), the municipality where serviceswere provided, provider ID, primary diagnosis, date ofservice, type of service, type of reimbursement (i.e., capi-tated, fee-for-service), units, payment by HMO, and patientcost-sharing (i.e., deductible, user fees). However, data didnot include information on the type of care provider (i.e.,general practitioner, neurologist) or procedures that werenot included in the government-defined benefitspackage. Second, an enrollment file was provided withinformation on patient´s ID, age, gender, home city, typeof enrollment, and educational attainment. Third, data ongovernment-paid services arrived with four fields: patientID, date of service, type of service (i.e., drugs, durablemedical equipment), and amounts reimbursed by the gov-ernment to the HMO. It is worth noting that all data wereaudited by the national government.

2.2. Defining PD cases using administrative data

To locate patients with PD from the data, we developed ana priori algorithm with increasing theoretical specificity for PD.First, to find probable PD cases, we selected patients fulfilling atleast one of the following criteria: at least one primary or sec-ondary hospital discharge diagnosis related to PD (InternationalStatistical Classification of Diseases (ICD)-10 G20, F02.3) anda minimum of one prescription of levodopa (LD) during theobserved period. Using these criteria, we selected 4,548 patients

(Figure 1). The primary ICD-10 diagnoses were chosen bya specialist of movement disorders, considering prior experi-ence, using administrative data [9]. Code F02.3 was added toinclude patients with dementia associated with PD. The use ofthis code allowed a better characterization of the complexityand progression of the disease.

Patients younger than 30 were excluded because of therare incidence of PD in this group [10–12]. To avoid possibleerrors associated with nonspecific ICD-10 codes, remainingsubjects (4,485 patients) were divided into two groups, asdiscussed below.

2.2.1. Patients included by ICD-10 codes or both ICD-10 andLD Anatomical Therapeutic Chemical Classification (ATC)codesTo increase algorithm specificity, if subjects were included only byICD-10 codes (G20 or FO2.3), the use of the same diagnostic codetwice or more was required to remain in the sample. Additionally,patients with at least one diagnosis and more than two anti-Parkinsonian drug prescriptions were included. All patients withfrequent drug-induced Parkinsonism prescriptions (DIPP) (see sup-plementary table) were excluded fromour sample. Figure 1-A showsdifferent exclusion criteria applied to avoid false positives.

2.2.2. Patients included only by LD anatomical therapeuticchemical classification (ATC) codesAlthough LD is the most effective dopaminergic therapy and hasthe greatest therapeutic consistency in all stages of PD, it can beused to treat other disorders, such as dystonia or restless legsyndrome. For that reason, patients taking LD associated withdystonia or other ICD-10 diagnoses were excluded. ‘Other ICD-10diagnoses’ include at least 50 ICD-10 codes different than Z000,G96.8, and G96.9, which are related to LD prescriptions. Finally,patients taking frequent LD doses associated with nonspecificcodes (i.e., Z000, G96.8, and G96.9) were excluded, owing to thelow probability of PD (see Figure 1-B). Frequent drug prescriptionwas defined as two or more prescriptions during the observedperiod, according to Chillag–Talmor et al. [13]

After this refinement, 3,450 subjects remained in the data-base. 3,103 were observed in 2014, 3,264 were observed in2015, and 2,917 were observed during both years. To avoidbias in expenditure estimations caused by enrollment changes(e.g., patients dying or dropping from the HMO in 2014 andthose newly diagnosed in 2015) in the population, we usedthe latter sample. All codes used and their definitions arelisted in the supplementary table.

2.3. Defining advanced and non-advanced PD usingadministrative data

Although patients with PD are mainly affected by motorsymptoms, there is a large spectrum of medical conditionsthat appear more frequently during advanced stages of thedisease. These conditions usually indicate PD progression andimpose a greater disability for the patient. Accounting for thelimitations of our methodology, we divided the sample intotwo groups (i.e., advanced and non-advanced PD) accountingfor the presence of motor and non-motor complications thatcould be identified using administrative data (ICD-10 and ATC

Article highlights

● Median annual all-cause direct cost to health systems was USD1,576● 30% of total annual direct cost is requested via the legal system● 96% of patients with PD had at least another chronic condition● About 16% of patients had advanced PD● Severity was associated with increasing average costs by USD3,416 (p

= 0.000)

2 S. I. PRADA ET AL.

codes). Advanced PD is usually characterized by severe motordeficits, very limited mobility, falls, and cognitive and psycho-tic problems [14–16]. Advanced PD occurs when the patientmeets at least one of the following criteria: the patient hasdementia; the patient has any of the following proceduresduring either of the 2 years (i.e., tracheostomy, gastrostomy,jejunostomy, or a deep-brain stimulator (DBS) implant); or thepatient uses a wheelchair.

● Advanced PD caused by dementia: dementia occurs in up to40% of patients with PD and is more frequent in advanceddisease stages [17]. Dementia occurs if the patient has atleast one of the diagnostic or prescription codes mentionedin the supplementary table. Because no local guidelinesrecommend the use of cholinesterase inhibitors for thepharmacological treatment of other cognitive impairments(e.g., mild cognitive impairment), it was assumed that allpatients taking these medications had dementia.

● Advanced PD caused by gastrointestinal dysfunction: themost commonly seen non-motor feature of PD is gastro-intestinal dysfunction. Regarding the use of gastrostomycodes, patients were included as ‘advanced PD,’ because,in rare instances, when other treatments failed and ade-quate nutrition could not be maintained, percutaneousendoscopic gastrostomy was occasionally needed. As

with other neurodegenerative diseases (ND), gastro-stomy is used, because patients suffer feeding and nutri-tion problems. These challenges are more frequent inadvanced stages of some NDs and are considereda criterion for determining disease severity [18].Although enteral feeding is not exclusive to NDs, thenumber of non-PD patients that require these type ofprocedure is low: 1.7% [19].

● Advanced PD caused by respiratory dysfunction: respira-tory dysfunction in PD is caused by a combination offactors, including restrictive changes, upper airwayobstructions, and abnormal ventilatory drive. In somesevere cases, tracheostomy may be needed.

● Advanced PD caused by implanted DBS devices: DBSsinvolve elective surgical procedures for patients withadvanced PD. DBSs are for patients with advanced PD, ascharacterized by the presence of motor fluctuations andinvoluntary movements associated with the response todopaminergic agents.

● Advanced PD caused by motor disability: owing to lim-ited mobility in advanced PD stages that could also becomplicated by postural instability, wheelchair use wasconsidered a criterion for advanced PD [20]. The use ofother codes was limited by the non-coding of someassistive devices and walking aids in the data.

Figure 1. Flow diagram of PD patients.Definitions: *Diagnosis without continuity: less than two diagnostic codes during the follow-up period. **Antiparkinsonian drugs: including levodopa, bromocriptine, biperiden oramantadine. ***Frequent prescription: two or more prescriptions of the medication during the follow-up period. ****DIPP.

EXPERT REVIEW OF PHARMACOECONOMICS & OUTCOMES RESEARCH 3

2.4. Statistical analysis and cost estimation

Actual payment to providers by HMOs was recorded in the data.Thus, monetary outlays were aggregated using a spreadsheet.We use the exchange rate of 2,372 COL per USD1 to convertfigures for both years to USD to allow for international compar-isons. The exchange rate is based upon the average for years2014–2015 as reported by Colombian Central Bank. Using thesame exchange rate for both years avoids changes in the directcosts of care caused by exchange rate fluctuations. In this case,our results are reported in constant dollars.

Health expenditure variables can be non-normally distribu-ted. Thus, Shapiro–Francia tests were performed to verifywhether samples were normally distributed. Medians with inter-quartile intervals were computed using Mann–Whitney tests onstatistical differences inmedians, applied to describe direct costs.For descriptive purposes, we broke down all-claims for all-condition cost for patients with PD by gender, age, stage (i.e.,advanced and non-advanced), type of expenditure (i.e., drugsand other services), and the number of chronic conditions. Thelatter was extracted after running the HMOs enrollment andclaims data through the Johns Hopkins adjusted clinical groups(ACG) System. The system uses ICD10 diagnosis codes and phar-macy ATC codes to assign enrollees to ACG and expandeddiagnosis clusters (EDC). The chronic condition count is a countof EDCs containing trigger diagnoses indicating chronic condi-tions with significant expected durations and resource require-ments. This measure has been widely used in the literature asa predictor of healthcare costs [21,22]. To further explore whatpart of total expense per person is attributable to advanced PDstages, we estimated a generalized linear model (GLM), witha link (log) function and a Poisson distribution supported bya Park test. The GLM was selected, because the total expensewas not normally distributed [23,24]. The dependent variable

was the cumulative annual expense for both years. To accountfor the confounding effect of other variables available in thedataset, we included as covariates, age, gender, and the interac-tion between age and gender, year, an indicator for HMO, anindicator for advanced PD, and an indicator for the city of resi-dence. Note, by including indicators of the city of residence, theestimated advance stage cost per patient adopts any time-invariant effect at the city- of residence-level of the patient.

3. Results

Total direct all-claims all-cause expenses for patients with PDin the selected sample in 2015 were USD10.7 million, out ofwhich, USD7.6 million was for technologies already covered bythe benefits package, and USD3.1 million was for technologiesclaimed via the judiciary system.

Table 1 gives median direct all-claims all-cause expensesper patient for patients with PD. Median direct costs increasedby 42% in 2015 (p = 0.000), from USD1,110 (IQR:USD2,548–410) to USD1,576 (IQR: USD3,617–605). By stage,the difference in medians between a patient with advancedPD and a non-advanced PD was USD1,774 in 2014 andUSD2,185 in 2015 (p = 0.000). The annual rate of growth waslower for advanced patients (30%) than for non-advancedpatients (42%; p = 0.000). Multimorbidity was highly prevalent,96% of PD patients had at least another chronic condition.

In Table 2, direct median all-claims-all-cause expenses perpatient are shown by the type of expenditure and whetherthey were covered by the national benefits package orclaimed via judiciary methods. Direct median costs of coveredprocedures per patient grew by 16% and of covered drugs by97%. The cost of technologies claimed using the legal systemgrew 420% across both years (p = 0.000).

Table 1. HMO direct per-capita expense for patients with PD in Colombia, 2014–2015 (USD).

2014 2015

N Median IQR [Q3-Q1] Median IQR [Q3-Q1] Median Change P-value

All 2,917 1,110 2,548-410 1,576 3,617-605 42% 0.000By GenderFemales 1,364 1,136 2,578-433 1,629 3,687-635 43% 0.000Males 1,553 1,066 2,524-386 1,550 3,569-595 45% 0.000By StageAdvanced PD 479 2,711 6,329–1,022 3,513 10,194–1,475 30% 0.000Non- Advanced PD 2,438 937 2,119-356 1,328 3,038-529 42% 0.000By Age Groups30–39 20 710 2,016-282 1,244 2,880-811 75% 0.09440–49 86 909 2,839-380 1,633 3,178-377 80% 0.42450–59 345 1,137 2,469-430 1,473 3,752-531 30% 0.00860–69 679 1,082 2,393-389 1,537 3,429-558 42% 0.00070–79 1,005 1,079 2,468-429 1,494 3,606-642 38% 0.00080–89 678 1,272 2,932-408 1,762 3,818-630 39% 0.000≥90 104 833 2,480-361 1,734 3,800-715 108% 0.001By Chronic Conditions0 107 324 1,449-105 701 1,812-164 116% 0.1441 604 661 2,076-213 854 2,524-294 29% 0.0082 712 800 2,255-324 1,181 2,952-436 48% 0.0003 588 1,120 2,288-472 1,492 3,728-586 33% 0.0004 392 1,278 2,742-672 1,895 4,802-856 48% 0.0005 253 1,692 3,616-860 2,052 4,058–1,030 21% 0.0446 149 1,810 3,777–1,121 2,311 5,066–1,349 28% 0.0277 o more 112 2,488 4,637–1,351 3,171 6,611–1,660 27% 0.030

Exchange rate $2,372 COP/1 USD. United States (US); Number of Patients (N); 25th percentile (Q1); 75th percentile (Q3); P-value: Prob > |z|-Mann–Whitney test (test on the equality of medians).

4 S. I. PRADA ET AL.

Lastly, results from regression analyzes showed that adjustedattributable costs to non-advanced PD were USD1,008, andthose to advanced PD were USD3,416 (p = 0.000).

4. Discussion

Ideally, a comprehensive cost-of-illness study on PD shouldinclude a societal perspective and longer time frames, becausePD often starts when people are still working [25]. According toBovolenta et al. [26], there were six types of Parkinson diseasecost studies published in the literature, depending on how costsare defined: direct medical, direct nonmedical, indirect cost,intangible, and personal. Direct cost includes hospitalization,pharmacotherapy, outpatient visits, diagnostics, and nursinghomes. The other type of studies included cost related to trans-port (direct nonmedical), loss of productivity (indirect), psycho-logical and psychosocial (intangible), and out-of-pocket(personal). Our study looked at the direct medical costs toHMOs in the context of a country with universal health coverage,a universal-government-defined benefits package, and univer-sal-government defined premiums to HMOs. A societal perspec-tive or a longer time frame was not the scope of this paper.Neither was possible owing to the lack of data.

We found that the median PD patients cost to the insurersin 2015 was USD1,576. We found an increase (42%; p = 0.000)in direct median cost-per-patient between the 2 years underanalyses. It is worth noting that these were all-cause claims.Therefore, patients’ direct costs to HMOs were associated withthe number of comorbidities, which may have worsened dur-ing the second year. We also found an increase of 420%(p = 0.000) in the direct median cost per-patient betweenthe 2 years for technologies outside the benefits packageclaimed via the judiciary system. It is an increment ofUSD212, accounting for about 13.4% of the total mediancost per patient in 2015. The percent change highlightsa worrisome issue. Requesting technologies using the judiciarysystem are inefficient and a burden for patients, judges, andinsurers.

There is no other study in Latin America reporting directmedian costs per-patient extracted from claims data. A studyconducted in Brazil by Bovolenta et al. [27] used a sample of 260self-selected patients who returned a questionnaire in whichinformation on PD-associated costs was requested covering theprevious 6 months, extrapolated to a full year. Cost data wascollected between October 2015 and September 2016. Directmean medical costs were USD1,511. However, after excluding

privately paid medical visits and medications, mean medicaldirect costs for the Brazilian national health system wasUSD1,271.

Our evidence shows that advanced PD is associated with anaverage increase per year per patient of around USD3,416 indirect costs. The same study from Brazil used the modifiedHoehn and Yahr disease severity scale, showing a positivecorrelation between costs and severity. For example, the dif-ference between patients having a score of 4 in the scale anda patient having a score of 1 was USD3,792.

According to the review by Bovolenta et al. [26], worldwidestudies have had difficulties obtaining large enough samples.Thus, most studies have fewer than 1,000 patients. With 2,917patients, our study is the largest in Latin American and theseventh-largest worldwide. Whereas this is one importantstrength of our study, there are limitations.

4.1. Limitations

As with any administrative data study, there are several limita-tions. Potential errors with the use of ICD-10 codes includeinaccurate diagnoses during early stages of the disease, miss-ing data, and coding errors. As reported by Szumski et al. [28],PD prevalence estimates may be particularly susceptible toerrors in coding because of the natural history of the diseaseand the lack of confirmatory diagnostic tests. Potential errorsand limitations of ATC codes include administration of thesame medications for other conditions, the absence of somedopaminergic agents in the HMO records, the use of levodopain PD diagnostic trials, including patients without the disease(controls), and inappropriate uses of these agents.

As awareness of the importance of having high-qualitydata has increased, more analyses have been published thatexplore diagnostic validity using a variety of approaches.We tried to minimize such errors in our algorithm by max-imizing the reliability of identifying appropriate patientswith PD and their outcomes while requiring that G20 andF023 be assigned at least twice and using only prescriptionsthat were frequent (i.e., 2+). Additionally, we excluded peo-ple in age groups having a low prevalence of the diseaseand patients with frequent prescriptions for medicationsassociated with secondary Parkinsonism. The use of G25.8and G25.9 codes may be problematic, given that somepatients with Parkinsonism can be formulated with levo-dopa. These codes are nonspecific and can mislead ouralgorithm. However, only 11.06% of our population was

Table 2. HMO direct per-capita expense for patients with PD by type of expenditure, 2014–2015 (USD).

2014 2015

All (n 2917) Median IQR [Q3-Q1] Median IQR [Q3-Q1] Change P-value

HMO covered servicesDrugs 98 419-31 193 489-69 97% 0.000Medical supplies* 0 21-0 0 34-0 0%Procedures 368 866-163 428 1,246-168 16% 0.000Total 651 1,641-281 791 2,060-362 21% 0.000Government covered servicesDrugs, supplies, procedures 50 787-0 262 1,227-0 420% 0.000Total 1,110 2,548-410 1,576 3,617-605 42% 0.000

25th percentile (Q1); 75th percentile (Q3); P-value: Prob > |z|-Mann–Whitney test (test on the equality of medians). * In both years the medianexpenditure for medical supplements was 0, for that reason percent change and median-difference test were not estimated.

EXPERT REVIEW OF PHARMACOECONOMICS & OUTCOMES RESEARCH 5

included by that criterion. Furthermore, the frequency oflevodopa prescriptions in our sample was relatively high(median 15, IQR 7–21) for the observed period, whichmakes the diagnosis of Parkinson’s less prone to error.

It is also possible that misclassification could bias the cost-of-care results. However, this could happen with newly diagnosedpatients. Per our exclusion criteria, they are not part of the study.Additionally, because PD patients may become high-cost, HMOauditing is stricter on these patients.

Regarding the definition of the complexity of the disease,although procedures such as tracheostomies are rare in PD,and respiratory dysfunction may be associated with otherclinical conditions, in our sample, only 13 patients had tra-cheostomies (0.39%). Of these, six had at least one othercomplexity indicator.

Although the use of administrative data has limitations, thestandardization of coding is amenable to screening by algo-rithms with appropriate inclusion criteria and sufficient restric-tions to confer high sensitivity and specificity for identification ofthe event of interest. However, information on the classificationcapacity of algorithms is limited in published studies. A studyfound that, compared to medical chart information, the electro-nic data on precisely definedmedical diagnoses were likely to bereasonably valid, contrary to those requiring confirmation bylaboratory tests [29].

Our study describes the characteristics of one particularhealthcare system, and results may not be generalizable toother populations or clinical settings because of differencesin coding patterns. Fundamental issues regarding datavalidity should be addressed in the use of administrativedata for epidemiological research. However, this representsa starting point, especially in countries with similar clinicalenvironments, and it also confirms the utility of applyingalgorithms based on the restrictions resembling the deci-sion-making process occurring in the clinical context. Ourresults are key findings and inputs for the economic evalua-tion and economic impact of PD in health systems in thedeveloping world.

5. Conclusion

This paper looked at HMO expenses associated withpatients with a PD diagnosis using administrative datainstead of survey data, separating expenses that were cov-ered under the universal health package and those thatwere provided on an individual basis via the judiciary sys-tem. Our results showed that median direct costs wereabove those of Brazil but lower than those of Europe. Weshowed that patients with PD must increasingly use thejudicial system to obtain healthcare services. This isa considerable burden on patients, the judicial system, andthe healthcare system. For Colombia, our study suggeststhat a thorough revision of the current package is necessaryso that these burdens can be made less onerous. For devel-oping countries where judicialization is an increasing phe-nomenon, to promote more equitable and efficient access,benefits packages should more thoroughly consider theneeds of these patients.

Acknowledgments

Sharon Cáceres for her excellent research assistance.

Funding

This paper was not funded.

Declaration of interest

The authors have no relevant affiliations or financial involvement with anyorganization or entity with a financial interest in or financial conflict withthe subject matter or materials discussed in the manuscript. This includesemployment, consultancies, honoraria, stock ownership or options, experttestimony, grants or patents received or pending, or royalties.

Reviewer Disclosures

Peer reviewers on this manuscript have no relevant financial or otherrelationships to disclose.

Author contributions

SP, AP, JV, MM, JO, and YT conceived and designed the research project.SP and AP performed the statistical analysis and wrote the first draft. SP,AP, JV, MM, JO, and YT reviewed and critiqued the manuscript. All authorsreviewed and agreed with the final manuscript.

ORCID

Sergio I. Prada http://orcid.org/0000-0001-7986-0959

References

Papers of special note have been highlighted as either of interest (•) or ofconsiderable interest (••) to readers.

1. Dorsey E, Elbaz A, Nichols E, et al. Global, regional, and nationalburden of Parkinson ’ s disease, 1990 – 2016 : a systematic analysisfor the global burden of disease study 2016. Lancet Neurol. 2016;17(11):939–953.

2. Rodríguez-Blázquez C, Forjaz M, Lizán L, et al. Estimating the directand indirect costs associated with Parkinson’s disease. Expert RevPharmacoecon Outcomes Res. 2015;15(6):889–911.

3. Von Campenhausen S, Winter Y, Rodrigues E Silva A, et al. Costs ofillness and care in Parkinson’s disease: an evaluation in six coun-tries. Eur Neuropsychopharmacol. 2011;21(2):180–191.

4. Low V, Ben-Shlomo Y, Coward E, et al. Measuring the burden andmortality of hospitalisation in Parkinson’s disease: A cross-sectionalanalysis of the English hospital episodes statistics database2009–2013. Parkinsonism Relat Disord. 2015;21(5):449–454.

5. Lökk J, Borg S, Svensson J, et al. Drug and treatment costs inParkinson’s disease patients in Sweden. Acta Neurol Scand.2012;125(2):142–147.

6. Tamás G, Gulácsi L, Bereczki D, et al. Quality of life and costs inParkinson’s disease: a cross sectional study in hungary. PLoS One.2014;9(9):e107704.

7. Findley LJ, Wood E, Lowin J, et al. The economic burden ofadvanced Parkinson’s disease: an analysis of a UK patient dataset.J Med Econ. 2011;14(1):130–139.

8. Prada SI, Soto VE, Andia TS, et al. Higher pharmaceutical publicexpenditure after direct price control: improved access orinduced demand? The Colombian case. Cost Eff Resour Alloc.2018;16(1):1–8.

9. Baldacci F, Policardo L, Rossi S, et al. Reliability of administrativedata for the identification of Parkinson’s disease cohorts. NeurolSci. 2015;36(5):783–786.

6 S. I. PRADA ET AL.

10. Hofman A, Collette HJ, Bartelds AI. Incidence and risk factors ofParkinson’s disease in the Netherlands. Neuroepidemiology. 1989;8(6):296–299.

11. Mayeux R, Marder K, Cote LJ, et al. The frequency of idiopathicParkinson’s disease by age, ethnic group, and sex in northernManhattan, 1988-1993. Am J Epidemiol. 1995;142(8):820–827.

12. Van Den Eeden SK. Incidence of Parkinson’s disease: variation by age,gender, and race/ethnicity. Am J Epidemiol. 2003;157(11):1015–1022.

• Relevant source for evidence for defining PD cases. This articleshowed that Patients younger than 30 had a rare incidence of PD.

13. Chillag-Talmor O, Giladi N, Linn S, et al. Use of a refined drug traceralgorithm to estimate prevalence and incidence of Parkinson’s dis-ease in a large Israeli population. J Parkinsons Dis. 2011;1(1):35–47.

•• This study served to refine our algorithm. In this article, PDprevalence was estimated using anti-parkinsonian drugs.Frequent drug prescription was defined as two or more pre-scriptions during the observed period.

14. Varanese S, Birnbaum Z, Rossi R, et al. Treatment of advancedParkinson’s disease. Parkinsons Dis. 2011;2010:1–9.

15. Poewe W, Seppi K, Tanner CM, et al. Parkinson disease. Nat Rev DisPrim. 2017;3:1–21.

16. Schapira AHV, Chaudhuri KR, Jenner P. Non-motor features ofParkinson disease. Nat Rev Neurosci. 2017;18(7):435–450.

17. Chaudhuri KR, Healy GD, Schapira H,VA. Non-motor symptoms ofParkinson’s disease: diagnosis and management. Lancet Neurol.2006;5(3):235–245.

18. Prada SI, Takeuchi Y, Merchán-Galvis AM, et al. Actual expenseassociated with patients with Alzheimer’s disease in Colombia.Int Psychogeriatr. 2017;29(11):1835–1840.

19. Wilmskoetter J, Simpson AN, Logan SL, et al. Impact of gastrostomyfeeding tube placement on the 1-year Trajectory of care in patientsafter stroke. Nutr Clin Pract. 2018;33(4):553–566.

20. Kaltenboeck A, Johnson SJ, Davis MR, et al. Direct costs and survivalof medicare beneficiaries with early and advanced parkinson’sdisease. Parkinsonism Relat Disord. 2012;18(4):321–326.

21. Corti MC, Avossa F, Schievano E, et al. A case-mix classificationsystem for explaining healthcare costs using administrative data inItaly. Eur J Intern Med. 2018;54:13–16.

22. Starfield B, Kinder K. Multimorbidity and its measurement. HealthPolicy. 2011;103(1):3–8.

23. Manning WG, Mullahy J. Estimating log models: to transform or notto transform? J. Health Econ. 2001;20(4):461–494.

24. Buntin MB, Zaslavsky AM. Too much ado about two-part modelsand transformation? Comparing methods of modeling medicareexpenditures. J Health Econ. 2004;23(3):525–542.

•• This study served to define our econometric model. This articlecompared different methods of modeling health expenditures.

25. Kowal SL, Dall TM, Chakrabarti R, et al. The current and projectedeconomic burden of Parkinson’s disease in the United States. MovDisord. 2013;28(3):311–318.

26. Bovolenta T, De Azevedo Silva S, Saba T, et al. Systematic reviewand critical analysis of cost studies associated with Parkinson’sdisease. Park Dis. 2017;2017:1–11.

•• This systematic literature review summarized cost studies onParkinson’s Disease.

27. Bovolenta T, De Azevedo Silva S, Saba T, et al. Average annual costof Parkinson’s disease in São Paulo, Brazil, with a focus on disease-related motor symptoms. Clin Interv Aging. 2017;12:2095–2108.

28. Szumski NR, Cheng EM. Optimizing algorithms to identifyParkinson’s disease cases within an administrative database. MovDisord. 2009;24(1):51–56.

29. Rawson NSB, D’Arcy C. Healthcare databases for drug safetyresearch: data validity assessment remains crucial. Drug Saf.2018;41(9):829–833.

EXPERT REVIEW OF PHARMACOECONOMICS & OUTCOMES RESEARCH 7