Trends in polypharmacy over 12 years and changes in its social gradients in South Korea

Polypharmacy is associated with adverse drug reactions and represents an economic burden on the health insurance system. The objective of our study was to assess the trends in polypharmacy and its associated factors in South Korea. This cross-sectional study used a nationwide sampled database between 2002 and 2013, including outpatients of all ages who received at least 1 prescription in the same period. Polypharmacy was defined as the concomitant prescription of ≥6 distinct medications on a single prescription at least once without a given duration. The yearly prescribing trends were calculated and plotted. We conducted comparative analyses to identify the changes in social gradients of polypharmacy between the first 2 years, 2002‒2003, and the final 2 years, 2012‒2013. We repeated logistic regressions for pediatrics <20 years of age and adults ≥20 years of age to estimate the adjusted odds ratios (aOR) and 95% confidence intervals (CI). The distributions of polypharmacy in the respective periods were examined according to patient economic status (0 = most deprived and 10 = most affluent). The age-standardized prevalence of polypharmacy decreased from 65.8% in 2002 to 43.7% in 2013. Our study included 1,108,298 outpatients throughout 2002–2013. Pediatric patients aged 1–9 years had the highest number of medications among all age groups (mean: 5.1 ± 1.1 in 2002–2003 vs. 4.1 ± 1.1 in 2012–2013) in both periods. Changes in the association between deprivation and polypharmacy over 10 years were observed in adults (aOR = 0.68; 95% CI = 0.62–0.75 in 2002–2003 vs. 1.60; 95% CI = 1.54–1.66 in 2012–2013) and pediatrics (aOR = 0.60; 95% CI = 0.52–0.68 in 2002–2003 vs. 1.07; 95% CI = 1.01–1.14 in 2012–2013) compared with those in the most affluent patients. The high level of polypharmacy in pediatric patients is a public health concern that warrants policymaker attention.


Introduction
Although appropriate polypharmacy based on the best evidence and optimized combinations of medicines may prolong a patient's life expectancy [1], inappropriate polypharmacy has the potential to cause adverse clinical outcomes. Polypharmacy increases the risk of drug-drug interactions (DDIs), which in turn trigger adverse drug reactions (ADRs) [2]. Polypharmacy was observed to be associated with ADRs, including falls, mortality, or dementia, which have a devastating effect on patients' quality of life [3][4][5][6][7][8][9]. Twenty-five percent of outpatients aged !18 years who received any prescription from participating primary care providers reportedly experienced an ADR, and their number of medications was significantly higher than that of the group without ADRs [10]. Multimorbidity is a well-known factor of polypharmacy and is largely associated with low socioeconomic status [11]. Patients with low socioeconomic status had an increased risk of mortality (risk ratio = 2.84; 95% confidence interval, CI = 2.25-3.60) [12], long working hours, and incident type 2 diabetes (risk ratio = 1.29; 95% CI = 1.06-1.57) [13]; a 2-fold increased risk of cardiovascular disease, coronary heart disease, and stroke [14]; and a 10-to 15-year earlier onset of multimorbidity [15]. Evaluating economic status is particularly important in polypharmacy research as it may affect multiple morbidity conditions and the consequent risk of polypharmacy [16].
Although most previous studies on polypharmacy focused on the elderly [8,17,18], evaluating polypharmacy in patients of all ages is essential for the following reasons. First, polypharmacy is largely driven by multimorbidity, and a significant number of people with multimorbidity are younger than 65 years, particularly in economically deprived populations [15]. Second, inappropriate medication overuse is highly prevalent in pediatric patients. Pediatrics younger than 9 years showed a more than 2-fold greater systematic antibiotic consumption than patients aged 10-59 years in South Korea [19]. Furthermore, 23% of pediatric ambulatory visits in the United States involved the inappropriate prescription of antibiotics for conditions that were unlikely to benefit from antibiotic treatment; this accounts for >10 million annual visits [20]. Thus, we included a population of all ages with the consideration of a potential earlier onset of multimorbidity in an economically deprived population, which in turn influences medication usage, and inappropriate medication overuse among pediatric patients.
Considering the increasing trends of polypharmacy in developed countries [21,22], identifying the time-series trends of polypharmacy is an important public health issue, and policy efficacy must be evaluated to reduce polypharmacy. Therefore, the primary objective of our study was to present the longitudinal time trend of prescriptions in South Korea from 2002-2013 using a large sample obtained from the health care insurance database that represented the entire population. Our secondary objective was to identify and compare the associated social health gradients with polypharmacy over 10 years.

Data source
We used the National Health Insurance Service-National Sample Cohort (NHIS-NSC), a population-based cohort established by the National Health Insurance Service (NHIS) in South Korea (Data number: NHIS-2018-2-024), and followed up from 2002-2013. After the initiation of the National Health Insurance program in Korea in 1977, it achieved universal coverage of the entire population in 1989. Accordingly, the database contains all information on health care use and prescribed drugs reimbursed by the NHIS for approximately 50 million Koreans. Information on non-reimbursed drugs or over-the-counter drugs is not covered by the database. The NHIS-NSC consists of approximately 1 million participants, corresponding to 2.2% of the entire Korean population, randomly selected from the NHI database [23]. The database includes demographic information on sex, level of income, and age groups, and clinical information on both inpatients and outpatients, drug prescriptions, and diagnoses of diseases coded based on the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM).

Study design and population
We carried out a cross-sectional study to estimate trends in polypharmacy over 12

Outcomes
Our study included all prescriptions in outpatient settings that were reimbursed by the South Korean NHIS. Medications included both regular and pro re nata (prn) medications. Polypharmacy was defined as the prescription of !6 distinct medications concomitantly on a single prescription at least once in our study. This prescription-based definition of polypharmacy was used in order to align our study with a policy implemented to reduce overuse and misuse of prescribed medications in South Korea. We selected this definition without a given duration because acute respiratory tract infection was one of the reasons for polypharmacy in South Korea [24]. We classified the patients into polypharmacy (!6 medications) and non-polypharmacy ( 5 medications) groups based on the maximum number of concomitantly prescribed medications. In the 12-year polypharmacy trend analyses between 2002 and 2013, the yearly prevalence of polypharmacy was calculated by dividing the sum of patients who received a polypharmacy prescription at least once by the total number of outpatients in the respective year. In the comparative analyses between the first 2 years (2002-2003) and the final 2 years (2012-2013), patients who received a polypharmacy prescription at least once during the respective period were defined to have an outcome.

Covariates
Multimorbidity was defined as the coexistence of 2 or more morbidities [15]. The definition of multimorbidity included the 51 most frequently reported chronic conditions by health care systems in Germany [25] and South Korea [26]. We used an unweighted count of morbidity as no single outcome was specifically selected for our purpose to describe the general morbidity status in the population. The conditions included physical comorbidities across system organ classes, including hypertension, diabetes, osteoporosis, chronic gastritis, liver disease, thyroid dysfunction, and osteoporosis, in addition to mental disorders, including dementia, anxiety, and depression (S1 File).
We used the latest records for individual patients in the first 2 years (2002-2003) and the final 2 years (2012-2013) to measure their age and level of income for the analyses. Age groups were divided into 10-year intervals ranging from 0-85 years or older to demonstrate the representative polypharmacy status of each generation in the graphs showing the distribution of polypharmacy and the logistic regression analysis.
Economic status was defined by the insurance fee, which is priced according to the level of income and properties, as assessed by the Korean NHIS. It consists of 11 deciles, ranging from 0 (the most deprived) to 10 (the most affluent). The 0th decile corresponds to medical aid patients and the most deprived population. The deciles are divided by 10% intervals ranging from the first to the tenth deciles in order, in which the first decile accounts for the lowest 10% and the tenth decile the highest 10%.

Statistical analysis
The prescription trends over 12 years between 2002 and 2013 were plotted and presented graphically. To assess the 12-year polypharmacy trends between 2002 and 2013, the yearly prevalence of polypharmacy was calculated by dividing the sum of patients who received a polypharmacy prescription at least once by the total number of outpatients in the respective year. As the distributions of age were heterogeneous across the years, the age-standardized prevalence of polypharmacy was calculated using the values of 2013, the last year of our study, as a reference. We calculated the prevalence of polypharmacy in each 10-year age interval of the standard population (2013). The expected number of patients exposed to polypharmacy in each year from 2002-2013 was respectively calculated across the age groups. We computed the standardized polypharmacy ratio using the sum of actual polypharmacy patients in each year as the numerator and the sum of expected polypharmacy patients as the denominator. The age-standardized prevalence of polypharmacy was calculated by multiplying the yearly prevalence by the calculated ratios.
The number of prescriptions and the mean number of medications were calculated on a yearly basis. The yearly changes in the mean number of medications per prescription and the age-standardized prevalence were plotted graphically. To examine the distributions in polypharmacy by age and income groups in the respective periods, we respectively plotted the association between polypharmacy and age in the first 2 years (2002-2003) and the final 2 years (2012-2013); the x-axis corresponded to the age in 10-year intervals and the y-axis is the mean number of medications. The plots were presented according to the economic status scale (0 = most deprived and 10 = most affluent). We calculated the mean medication use for each patient.
Patient characteristics, including age, sex, level of income, multimorbidity, CCI, and comorbidities, were summarized as counts with proportions for categorical variables and means with standard deviations (SD) for continuous variables between the polypharmacy and non-polypharmacy groups. Chi-square tests were used to test for significant differences between groups.
We conducted a logistic regression analysis to calculate the odds ratios (aOR) and 95% CIs of the associations between patient characteristics and polypharmacy. We repeated the logistic regression analyses for pediatrics and adolescents (<20 years of age) and adults and the elderly (!20 years of age) in the first 2 years (2002-2003) and the final 2 years (2012-2013). Models were adjusted for sex, age group, level of income, and multimorbidity. To avoid multicollinearity, we excluded the following variables from the logistic model: comorbidity and CCI, which overlapped with multimorbidity, and health insurance type, a surrogate variable for the level of income. Aside from multimorbidity, comorbidity comprised of 7 highly prevalent morbidities: hypertension, diabetes, hyperlipidemia, ischemic heart disease, cerebrovascular disease, arterial disease, and kidney disease. We carried out a likelihood ratio test for the estimation of the p-for-trend between economic status and multimorbidity. All statistical procedures were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

Sensitivity analysis
We applied a threshold for polypharmacy of as !5 simultaneously prescribed medications in a single prescription to evaluate the robustness of the results of polypharmacy trends. To assess the weight of cancer patients contributing to the prevalence of polypharmacy, we limited study subjects to non-cancer patients in the sensitivity analysis. Patients with diagnostic records of cancer (ICD-10: C00-C97) in the outpatient setting within the respective year were excluded from the analysis. The polypharmacy trends in non-cancer patients were presented with the yearly age-standardized prevalence and the respective prevalence in pediatrics and adolescents (<20 years of age) and adults and the elderly (!20 years of age) across 2002-2013.
As the results may have been influenced by two extreme categories (most deprived compared to most affluent groups on a 10-item scale), we conducted a sensitivity analysis with patients' insurance type in the NHIS, either medical aid recipients (0th decile) or National Health Insurance beneficiaries (1st decile-10th decile), instead of using income deciles. In this analysis, the insurance type, either medical aid recipient or National Health Insurance beneficiary, was binary coded, and the logistic model was adjusted for sex, insurance type, and multimorbidity.

Results
Our study observed decreasing trends in polypharmacy for 1,108,298 outpatients (Fig 1). The prevalence of polypharmacy and the mean number of medications showed a decreasing trend (Fig 2). The age-standardized prevalence decreased from 65.8% in 2002 to 43.7% in 2013; that of pediatrics and adolescents (<20 years of age) decreased from 65.3% to 46.6% and that of adults and the elderly (!20 years of age) decreased from 48.4% to 42.9% (Table 1) Fig 3); moreover, infants aged 0 years had a significantly lower mean number of medications. However, the most deprived patients had a higher mean number of medications than other deciles in 2012-2013 (Fig 4).
In the logistic regression analysis of polypharmacy status (Table 3), we observed discrepant findings in the association between the level of income and polypharmacy in the 2 study periods. In 2002-2003, the most deprived participants had a decreased risk of polypharmacy compared with the most affluent participants in both pediatrics and adolescents (aOR = 0.60; 95% CI = 0.52-0.68) and adults and the elderly (aOR = 0.68; 95% CI = 0.62-0.75). Conversely, we observed a non-significant risk in pediatrics and adolescents (aOR = 1.07; 95% CI = 1.01-1.14) and an increased risk in adults and the elderly (aOR = 1.60; 95% CI = 1.54-1.66) in 2012-2013. The odds of polypharmacy generally increased with multimorbidity across study periods and age groups ( Table 3). The trends between economic status and multimorbidity were statistically significantly different with a P-value of <0.05 (results not shown). The directions of the association were consistent in the sensitivity analysis using insurance type, binary coded for

Discussion
Our study was designed to present long-term polypharmacy trends, their distributions, and changes in associating factors over 10 years. This real-world evidence showed gradually decreasing trends in polypharmacy over 12 years. Pediatric patients aged . This decrease reflects efforts to reduce prescriptions made by the Korean Public Health Authority, which adopted the Evaluation Project on Appropriate Prescribing (EPAP) in 2001 to reduce the misuse and abuse of prescription medicines by improving the autonomous management of health institutions [28]. Apart from the decreasing trends in polypharmacy, the observed increase in the number of prescriptions may be the result of the aging of the study cohort.
Our finding of a decreasing trend in polypharmacy is contrary to that of other studies that used a single prescription-based definition and reported an increase in polypharmacy over time. A study in the United States reported that the prevalence of polypharmacy (!5 medications) rose from 8.2% to 15% overall between 1999 and 2012 and from 24% to 39% in participants aged !65 years [29]. An increase was also observed in Scotland among those !20 years of age, from 11.4% to 20.8% between 1995 and 2010 (!5 medications) [30], and from 24.7% to 31.6% in New Zealand between 2005 and 2013 among elderly people aged !65 years [31]. This conflict in trends may be due to the high level of polypharmacy in the Korean population. South Korea had more drug items per prescription (mean: 4.16) than the United States (mean: 1.97), United Kingdom (mean: 3.83), Japan (mean: 3.00), France (mean: 4.02), and Germany (mean: 1.98) in 2005 [24]. One reason for the very high number of drugs on a single prescription in Korea may be the high use of medications for acute upper respiratory infection (mean: 4.73) [24]. Although a decreasing trend in polypharmacy was observed in South Korea, the magnitude of the prevalence remained high.
Of note is that the pediatric patients 1-9 years of age had the highest number of medications among all age groups. This is well-supported by a finding that children 6 years of age had a higher level of antibiotic consumption than adults in South Korea [32], while the reverse was found in other Organization for Economic Co-operation and Development (OECD) countries [24]. Moreover, infants in Korea had the highest antibiotic use among 6 countries (Germany, Italy, South Korea, Norway, Spain, and the United States) [33]. The medications most frequently associated with pediatric patients were treatments for the common cold throughout the first years of our study, 2002-2003, and the recent years, 2012-2013 (S6 File). Although our study observed a decreasing trend in polypharmacy in pediatrics <20 years of age from 65.3% in 2002 to 46.6% in 2013, antibiotic consumption was still prevalent in this population in the last years of our study. Our results suggest the need for subsequent studies focusing on medication use in pediatrics and prescription guidelines in clinical practice to reduce polypharmacy in the population.
The highest medication consumption in pediatric patients is discordant with the positive association between the number of prescription medications and increasing age observed in the United States [34], where pediatrics aged 18 years had the lowest proportion of 5    prescription drugs among all age groups, and in Canada [35], where the proportion of 5 prescription drugs increased with age. Polypharmacy in pediatric patients could be particularly problematic considering their physical vulnerabilities [36,37] and higher risk of potential DDIs owing to the differences in their response to drugs compared with adults [38]. Pediatrics can react differently to drug administration as their pharmacokinetic processes of absorption, distribution, metabolism, and excretion are different from those of adults [39]. Contrary to our hypothesis that deprivation accelerates multimorbidity and its consequent polypharmacy, our results showed associations between deprivation and polypharmacy independently from multimorbidity. Adults in the most deprived group had a 40% decreased risk of polypharmacy compared with the most affluent in 2002-2003, whereas the most deprived had a 60% increased risk of polypharmacy in 2012-2013. This reversed association over 10 years may be attributed to the decrease in medical utilization among the reference group, the most affluent, due to differences in the cost-sharing policy in South Korea between the study periods. This may have influenced the relatively higher medical utilization among the most deprived patients, and consequently increased the odds of polypharmacy after the policy reform. The cost-sharing policy was reformed in 2007 from copayment to coinsurance for National Health Insurance beneficiaries younger than 65 years, with the exclusion of medical aid patients corresponding to the 0th decile, i.e., those who are the most deprived. In order to reduce the burden of rapidly increasing health expenditures after the initiation of the National Health Insurance program in 1989 [40], a patient cost-sharing change was implemented for outpatients aged younger than 65 years on August 1, 2007. In the copayment system, outpatients had a fixed medical expense of KRW (Korean won) 3,000 from their own money unless the total costs exceeded the threshold of KRW 15,000; if the total costs were greater than the threshold, patients were obligated to pay 30% [41]. After the coinsurance policy was implemented, national health beneficiaries (1st decile-10th decile) shared 30% of the total costs from their own money, regardless of total costs [42], whereas the copayment system was still applied to medical aid patients. The NHIS estimated that the average out-of-pocket cost for primary care visits for the common cold increased from KRW 3,110 to KRW 3,300 after the reform and prospected that the burden on the NHIS fund would be reduced by USD 193 million, annually [43].
In 2012-2013, after the change to the coinsurance system, decreased health care utilization among the most affluent patients may have influenced the relatively higher health care utilization in the most deprived group, resulting in an increased OR. Our interpretation is in line with a previous finding that medical care utilization, defined as the proportion of all beneficiaries in each group who visited clinics and the mean number of visit days per beneficiary, decreased after the cost-sharing change to coinsurance in South Korea [41]. Moreover, medical accessibility for the most deprived participants was enhanced by the coverage expansion of the Medical Aid Program to patients with rare, intractable, and chronic diseases, as well as children under the age of 18 in 2004 [44]. The increased association between medical aid recipients and polypharmacy in 2012-2013 also corresponds with a South Korean study that estimated a 52% increased risk of polypharmacy among elderly outpatients receiving medical aid in 2010-2011 [45], although the age inclusion criterion was different. Our study has several strengths. We generated generalizable results by using a representative electronic health insurance database. In addition, we determined the trends in polypharmacy between 2002 and 2013 using longitudinal data over a 12-year period. This is particularly important evidence to evaluate the outcome of South Korea initiating the EPAP in 2001 [28] to reduce the misuse or overuse of prescription medications. To the best of our knowledge, this is the first study presenting long-term trends for polypharmacy in a population of all ages in South Korea. By evaluating polypharmacy in a population of all ages, we identified that pediatric patients aged 1-9 years had the highest medication use. The economic status used in our study was objective and trustworthy as it was assessed by the Korean Public Health Authority according to the level of income and private properties, including real estate and vehicles. The monthly contribution is calculated by the monthly average wage multiplied by contribution rate for employee health insurance subscribers, and by the contribution score multiplied by the unit price for those with self-employed health insurance [46].
However, our study should be interpreted with caution owing to the following limitations. The definition of polypharmacy was based on the maximum number of concomitant medications on a prescription without a given duration. The prescription-based definition of polypharmacy might lead to an underestimation of polypharmacy by failing to capture doctorshopping behavior. For example, patients who were simultaneously prescribed with <6 medications in multiple primary care institutions will be classified in the non-polypharmacy group even if the sum of prescribed medications for the different institutions exceeded !6, the threshold of polypharmacy. This definition may also lead to an overestimation of polypharmacy by the assessment without a given duration of time. For example, patients with shortterm polypharmacy prescriptions will be classified in the polypharmacy group. A definition reflecting the duration of prescriptions may have been more appropriate for the medications prescribed for chronic disease. It is particularly problematic when comparing the results with studies using a duration-based definition. Discrepancies in the definition of polypharmacy were observed; for example, some studies defined polypharmacy as at least 5 medications [47,48], 6 medications [49,50], 10 medications [51], or used operational definitions, such as polypharmacy, that continued for more than a given duration of time [52][53][54].
However, for our objective to capture all prescriptions in primary care, we judged this prescription-based definition of polypharmacy to be more appropriate. Hence, the definition enabled our study to capture the medications prescribed for a short duration that are largely overused in the outpatient setting in South Korea, such as antibiotics [55], analgesics, and nonsteroidal anti-inflammatory drugs [56]. Our polypharmacy definition was aligned with the EPAP indicator in South Korea, a policy to reduce the misuse and abuse of prescription medicines. For polypharmacy, the EPAP indicator was defined as !6 drugs prescribed concomitantly on 1 prescription [28,57]. To evaluate longitudinal polypharmacy trends and the efficacy of policies, we aligned our polypharmacy definition with the policy. Moreover, the majority of former polypharmacy studies used a prescription-based definition without a given duration of time [58]. According to a systematic review exploring the definitions of polypharmacy, only 9 studies considered a given duration of time among 99 included studies [58]. Thus, we judged the prescription-based definition of polypharmacy to be more appropriate to increase comparability with previous findings.
The limitation of excluding non-reimbursed medications or over-the-counter drugs may lead to an underestimation of polypharmacy. However, this is not a significant hurdle as our study's objective was to measure polypharmacy of drugs reimbursed by the health care system. Furthermore, problems in interpreting the results may occur by combining multilevel variables in the multivariate logistic regression analyses. Our outcome, polypharmacy, involves multiple stakeholders, including individual patients, medical institutions, and health care systems, which is contrary to individual-level independent variables, such as age, sex, income, and multimorbidity. Other important unmeasured factors that may have an association with polypharmacy, such as alcohol consumption [59] or smoking status [21], were not controlled for in the analyses. Finally, the limitation of using prescription data underlies our study, as we do not know if patients actually took the medications prescribed.
In conclusion, our study observed that the prevalence of polypharmacy has declined over 12 years in South Korea. We provided evidence demonstrating that the highest risk of polypharmacy across all age groups was in pediatric patients 1-9 years of age. Considering the risk of polypharmacy and the vulnerability of pediatric patients, high medication use in this population is an important public health concern that should warrant the attention of policymakers and clinicians, as well as further research. Implementation of strategies to manage polypharmacy will be critical for safe and quality care in this population.