==== Front Sci Rep Sci Rep Scientific Reports 2045-2322 Nature Publishing Group UK London 78236 10.1038/s41598-020-78236-y Article Targeting continuity of care and polypharmacy to reduce drug–drug interaction Weng Yi-An 1 Deng Chung-Yeh 2 Pu Christy cypu@ym.edu.tw 1 1 grid.260770.40000 0001 0425 5914Institute of Public Health, National Yang-Ming University, 155 Li-Nong ST. Sec 2, Peitou, Taipei, Taiwan 2 grid.260770.40000 0001 0425 5914Institute of Hospital and Healthcare Administration, National Yang-Ming University, 155 Li-Nong ST. Sec 2, Peitou, Taipei, Taiwan 4 12 2020 4 12 2020 2020 10 212796 8 2020 19 11 2020 © The Author(s) 2020Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.Drug–drug interaction (DDI) is common among the elderly, and it can have detrimental effects on patients. However, how DDI can be targeted has been under-researched. This study investigates whether DDI can be reduced by targeting continuity of care (COC) through reducing polypharmacy. Population claims data of Taiwan National Health Insurance were used to conduct a 7-year-long longitudinal study on patients aged ≥ 65 years (n = 2,318,766). Mediation analysis with counterfactual method and a 4-way decomposition of the effect of COC on DDI was conducted. Mediation effect through excessive polypharmacy differed from that through lower-level polypharmacy. Compared with the low COC group, the high COC group demonstrated reduced excess relative risk of DDI by 26% (excess relative risk =  − 0.263; 95% Confidence Interval (CI) =  − 0.263 to − 0.259) to 30% (excess relative risk =  − 0.297; 95% CI =  − 0.300 to − 0.295) with excessive polypharmacy as the mediator. The risk only reduced by 8% (excess relative risk =  − 0.079; 95% CI, − 0.08 to − 0.078) to 10% (excess relative risk =  − 0.096; 95% CI, − 0.097 to − 0.095) when the mediator was changed to lower-level polypharmacy. The effect of COC on DDI was mediated by polypharmacy, and the mediation effect was higher with excessive polypharmacy. Therefore, to reduce DDI in the elderly population, different policy interventions should be designed by considering polypharmacy levels to maximize the positive effect of COC on DDI. Subject terms Health careRisk factorshttp://dx.doi.org/10.13039/501100004663Ministry of Science and Technology, Taiwan109-2314-B-010 -049 -MY2Pu Christy issue-copyright-statement© The Author(s) 2020 ==== Body Introduction Drug–drug interaction (DDI) events are common in the elderly population. The prevalence of clinically relevant DDI may be as high as 28.1% in the elderly population1. However, DDI-related adverse reactions are generally predictable and amendable2. Many avoidable hospital admissions for drug-related toxicity were due to preventable DDIs3. Continuity of care (COC) comprises 2 elements: how patients experience integration of services and how care is delivered over time4. In the recent years, enhancing patient COC has become one of the central themes in patient care. Greater COC is associated with fewer emergency department visits and hospital admissions and lower health care costs, which can be partially explained by the reduction of inappropriate medication use5. Higher COC is also related to better medication adherence and a reduction in drug-related problems such as inadequate dosing and inappropriate regimens6,7. High COC can reduce DDI event frequency, and this effect is greater in population with multiple comorbidities8. Despite its medical benefits, research on the mechanism through which high COC acts as a protective factor against DDI incidences remains insufficient. Knowledge on the mechanism of how COC affects DDI is essential for 2 reasons. First, ignoring the mediation effect can likely lead to biased estimates and render study results invalid. Second, mediation analysis allows policies to maximize the effects of COC on DDIs through the intervention of a mediator and hence improve an intervention’s components. Such studies on COC-DDI relationship are thus scant. Polypharmacy, which refers to the simultaneous usage of multiple medications, has been found to be a risk factor for DDI in older people1,2. DDI events can occur even when all the prescription guidelines are followed9. With increasing number of comorbidities and physicians involved, polypharmacy has become an inevitable consequence for patients10. The greater prevalence of multi-comorbidities in older people makes polypharmacy a common phenomenon for this population9. In an outpatient setting, 39.4% of the elderly population were exposed to polypharmacy11. Managing polypharmacy has been an issue in patient care in the elderly population. Increasing COC reduces the number of medications. A retrospective study demonstrated that both physician continuity and site continuity exert protective effect over medication duplication, and the effect was greater for elderly patients with multi-comorbidities12. A Japanese study discovered that an older patient without a regular physician was 2.5 times more likely to experience polypharmacy than those who had one13. The association between greater COC and positive patient outcomes has been well studied and investigated in many settings. However, the mechanism through which COC affects patient outcomes and reduces DDI has not been investigated. Polypharmacy is a DDI risk factor that is not only influential but also identifiable and amendable. Identifying its role in patient care is essential for elderly patients. The relationship between COC, polypharmacy, and DDI must be investigated to improve patient outcomes for elderly people. The results of our study thus possess policy relevance for the improvement of elderly patients’ outcomes. In addition to possessing policy relevance, this study addresses a shortcoming of other research (i.e., failure to use rigorous statistical methods for estimating mediation effects). This study tested whether the effect of COC on DDI is mediated by polypharmacy, and if so, what is the magnitude of the mediating effect. Methods In this study, we used the Taiwan National Health Insurance (NHI) claims database, managed by Taiwan’s Ministry of Health. The NHI program has been implemented in Taiwan since 1995 and is compulsory for all citizens. To date, the enrollment rate has reached 99.6% of the population. All the medical records from both inpatient and outpatient settings, including all prescriptions and procedures, are included in the database. The NHI claims data have been used frequently for studies involving drugs and pharmaceuticals because it provides accurate information on prescriptions, which is generally absent in the survey data generally5,6,14. This study was approved by the Institutional Review Board of National Yang-Ming University. Inclusion criteria Here, the elderly population was defined as people who are ≥ 65 years, and all records from the eligible patients’ outpatient visits were extracted. Patients who met the following inclusion criteria were enrolled: (1) > 1-year-long survival since 2011 and (2) ≥ 3 physician visits in the same year during the study period. These criteria were necessary as the COC index (COCI) can only be calculated within a time interval and with a reasonable number of visits15. By using these inclusion criteria, a total of 2,318,766 people were included in 2011. This cohort was then studied for 7 years until 2017. Exposure: COC We used the COCI to measure care continuity. This measurement is more suitable for health systems with a high number of outpatient visits16. COCI measures the concentration of visits in a certain period17. Physician COCI is thus calculated as follows18: PhysicianCOCI=∑i=1pni2-NN(N-1), where ni is the number of visits to the given physician i, p the number of the physicians that the patient visited, and N the total number of physician visits. Mediator and outcome variables To analyze polypharmacy and DDI, only Western medicine and drugs were included. Drugs that required only topical application, Chinese herbal medications, and over-the-counter medications were excluded. To identify the medication prescribed for each patient, we applied the Anatomical Therapeutic Chemical code as the classification index. Studies have used various cutoff points for the number of medications to define polypharmacy19. However, most studies have used 5 medications as the cutoff points11,20,21; hence, we also employed this cutoff for our study to ensure comparability. Thus, patients who were simultaneously on ≥ 5 medication types in a 3-month window were considered to be having polypharmacy; those who were being administered ≥ 10 medication types were defined as having excessive polypharmacy14,22. DDIs can be examined using different approaches, such as by using knowledge bases (KBs)2,23 and using the system used for clinical evaluation in a single country24,25. To account for the clinical severity of DDIs, professionals were recruited for evaluating the harm to the human body26,27. This study used a list of 15 types of DDIs identified as clinically significant by Phansalkar et al.28; this list has been adopted by several studies and in practice8,29,30. We selected this list as our target DDIs for the following reasons: (1) the information on the DDIs was cross-compared with the different KBs, (2) the list was strictly reviewed by a panel of experts from different backgrounds who had reached the consensus that the patients were critical and should be integrated into the clinical decision support system, and (3) the list presented the drug-class level of the DDI pairs, which allowed us to link it with the information we had in our data set. Some modifications were made to the list to exclude those combinations that were not covered by insurance or were not applied to the population. Only 10 types of DDIs were consequently included in the subsequent analysis. A DDI event was defined as 2 target drugs of a DDI pairs that were prescribed, with treatment periods that overlapped for at least 1 day. Statistical analysis To test the hypothesis that polypharmacy acts as a mediator between COC and DDI, we applied a counterfactual method because the exposure and mediator were measured simultaneously31. The product method and the difference method are the 2 traditional methods for mediation analysis. The difference method entails a comparison between the original effect of the exposure on the outcome and the one adjusted for the mediator, and the product method estimates the mediation effect by multiplying the effect of the exposure on the outcome and the effect of the mediator on the outcome after adjustments for the exposure and other covariates32,33. The traditional methods have been criticized for their lack of causal interpretation that leads to a biased estimation of the indirect effect with incorrect conclusion for mediation when applying a logistic regression in nonrare outcomes or with an exposure-mediator interaction term34,35. We used counterfactual method with 4-way decomposition for the mediation analysis36, which is an advanced method that decomposes the overall effect into 4 components: controlled direct effect (CDE), reference interaction (INTref), mediated interaction (INTmed), and pure indirect effect (PIE). Two regression models are required for the estimation: (1) the outcome model, a function of the exposure, mediator, and interaction between theses and the confounders, and (2) the mediator model, a function of the exposure and confounders37. Logistic regressions were estimated for both models. We defined high and low COC groups on the basis of the median of the COCI. DDI events and polypharmacy status were treated as dichotomous variables (with or without DDIs or polypharmacy). For sensitivity analysis, we redefined COC, polypharmacy, and DDI to determine whether our results were robust to different specifications. For COC, we recalculated the COCI using 2 visits, consistent with a previous study38. For defining polypharmacy and excessive polypharmacy, we used a longer time window, such that if a patient had received ≥ 5 medications repeatedly and persistently for > 181 days, it would be defined as persistent polypharmacy39. This alternative time window was also applied to define persistent excessive polypharmacy22. DDI was defined as the presence of a combination for ≥ 2 days (rather than 1 day) to ensure that our results were robust because of the stricter definition. We also regrouped COC by using tertiles to create low, intermediate, and high COC groups, which was consistent with previous studies5,40. For subgroup analysis (results not shown), we selected the 2 demographic characteristics that had been confirmed to have impact on COC and polypharmacy. A Korean retrospective study that focused on patients aged ≥ 65 years with diabetes, hypertension, asthma, and chronic obstructive pulmonary disorder found that COC was higher in men than in women and tended to decrease with age40. The differences between men and women in terms of the attitude toward medications and doctor-seeking behaviors were significant. Studies have revealed that women have higher physicians visit rates and have more positive attitudes toward medication therapies than men do41,42, and older people are more likely to receive polypharmaceutical therapies than younger people do42. All statistical analyses were conducted using STATA MP Environment (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC.). Results Table 1 lists the characteristics of the study population at baseline categorized by low and high COC groups. Significant differences were observed for sex and age between the 2 groups (P < .001). The number of physician visits and prescribed medications and the proportion of patients with polypharmacy and DDI events were significantly lower in the high COC group than in the low COC group (P < .001). During the study period, there was a little fluctuation in the means of the COCI. The changes of the proportions of polypharmacy cases and excessive polypharmacy cases were little. However, the percentage of DDI events in the study population increased over the years.Table 1 Baseline patient characteristics. Variables Low COC group High COC group P valuea All patients N 1,159,069 1,159,697 2,318,766 COCI [mean(SD)] 0.15(0.06) 0.46(0.21)  < 0.001 0.30(0.22) Age in 2011 [mean(SD)] 74.71(6.94) 74.72(6.96) 0.382 74.71(6.95) Age group [n (%)] < 0.001 65–74 629,334(54.30) 634,364(54.70) 1,263,698(54.50) 75–84 412,402(35.58) 406,122(35.02) 818,524(35.30) 85+ 117,333(10.12) 119,211(10.28) 236,544(10.20) Gender [n (%)] < 0.001 Male 532,303(49.26) 548,359(50.74) 1,080,662(46.61) Female 626,766(50.62) 611,338(49.38) 1,238,104(53.39) Physician visits [mean(SD)] 35.05(23.44) 25.25(17.91) < 0.001 30.15(21.42) Physician visits frequency [n(%)] < 0.001 Low 306,827(26.47) 487,507(42.04) 794,334(34.26) Intermediate 347,785(30.01) 422,063(36.39) 769,848(33.20) High 504,457(43.52) 250,127(21.57) 754,584(32.54) No. of prescribed medications [mean(SD)] 11.29(6.41) 8.17(4.81)  < 0.001 9.73(5.88) Polypharmacy status [n(%)] < 0.001 No 66,193(5.71) 141,473(12.20) 207,666(8.96) Yes 1,092,876(94.29) 1,018,224(87.80) 2,111,100(91.04) Drug–drug interaction events [n(%)]  < 0.001 No 1,104,885(95.32) 1,135,212(97.88) 2,240,097(96.60) Yes 54,184(4.67) 24,485(2.11) 78,669(3.39) COCI continuity of care index, SD standard deviation. aChi-square tests were used for comparisons between age group, gender, physician visits frequency, polypharmacy status, and DDI events, and t tests were used to compare COCI, physician visits, and number of prescribed medications. Tables 2 and 3 present the result of the 4-way decomposition for the relationship between COC, polypharmacy or excessive polypharmacy, and DDI. These estimates represent the excess relative risk of DDI due to the total effect, CDE, INTref, INTmed and PIE, signifying a change in the risk percentage compared with the reference group. The CDE was estimated at a fixed level of the mediator. In Table 3, polypharmacy was treated as the mediator, and the total excess relative risk was estimated to be − 0.565 (P < .001) in 2011, indicating that the cases of DDI events decreased by 56.5% in the high COC group than in the low COC group. Similar results were found when we used excessive polypharmacy as the mediator; the total excess relative risk was − 0.570 (P < .001; Table 3). The excess relative risk of COC on DDI events due to the PIE (pure mediation) of polypharmacy was significant: the one due to polypharmacy ranged from − 0.079 to − 0.096, and the one due to excessive polypharmacy was greater, ranging from − 0.261 to − 0.297 (both P < .001). The excess relative risk is attributable to reference interaction (interaction only) and mediated interaction (mediation and interaction); however, both of them slightly increased the excessive risk of COC on DDI events.Table 2 Four-way decomposition of the relationship between COC, polypharmacy, and DDI. 2011 2012 2013 2014 2015 2016 2017 βa,b p β p β p β p β p β p β p Mediator at 0 Total excess RR − 0.565 < 0.001 − 0.576 < 0.001 − 0.581 < 0.001 − 0.602 < 0.001 − 0.619 < 0.001 − 0.612 < 0.001 − 0.622 < 0.001 Excess RR due to controlled direct effect 0.004 0.193 0.004 0.131 0.008 0.003 0.001 0.878 0.007 0.020 0.003 0.251 0.005 0.112 Excess RR due to reference interaction − 0.533 < 0.001 − 0.546 < 0.001 − 0.551 < 0.001 − 0.564 < 0.001 − 0.589 < 0.001 − 0.578 < 0.001 − 0.589 < 0.001 Excess RR due to mediated interaction 0.043 < 0.001 0.043 < 0.001 0.048 < 0.001 0.052 < 0.001 0.055 < 0.001 0.053 < 0.001 0.057 < 0.001 Excess RR due to pure indirect effect − 0.079 < 0.001 − 0.078 < 0.001 − 0.086 < 0.001 − 0.090 < 0.001 − 0.092 < 0.001 − 0.091 < 0.001 − 0.096 < 0.001 Mediator at 1 Total excess RR − 0.565 < 0.001 − 0.576 < 0.001 − 0.581 < 0.001 − 0.602 < 0.001 − 0.619 < 0.001 − 0.612 < 0.001 − 0.622 < 0.001 Excess RR due to controlled direct effect − 0.568 < 0.001 − 0.579 < 0.001 − 0.583 < 0.001 − 0.602 < 0.001 − 0.621 < 0.001 − 0.612 < 0.001 − 0.620 < 0.001 Excess RR due to reference interaction 0.039 < 0.001 0.038 < 0.001 0.040 < 0.001 0.039 < 0.001 0.039 < 0.001 0.036 < 0.001 0.036 < 0.001 Excess RR due to mediated interaction 0.043 < 0.001 0.043 < 0.001 0.048 < 0.001 0.052 < 0.001 0.055 < 0.001 0.053 < 0.001 0.057 < 0.001 Excess RR due to pure indirect effect − 0.079 < 0.001 − 0.078 < 0.001 − 0.086 < 0.001 − 0.090 < 0.001 − 0.092 < 0.001 − 0.091 < 0.001 − 0.096 < 0.001 RR relative risk, COC continuity of care, DDI drug–drug interaction. aβ: model estimates. bLow COC group is the reference group. Table 3 Four-way decomposition of relationship between COC, excessive polypharmacy, and DDI. 2011 2012 2013 2014 2015 2016 2017 βa,b p β p β p β p β p β p β p Mediator at 0 Total excess RR − 0.570 < 0.001 − 0.581 < 0.001 − 0.586 < 0.001 − 0.606 < 0.001 − 0.623 < 0.001 − 0.616 < 0.001 − 0.625 < 0.001 Excess RR due to controlled direct effect 0.001 0.767 − 0.002 0.635 − 0.001 0.794 − 0.011 0.002 − 0.010 0.004 − 0.015 < 0.001 − 0.019 < 0.001 Excess RR due to reference interaction − 0.437 < 0.001 − 0.449 < 0.001 − 0.444 < 0.001 − 0.453 < 0.001 − 0.475 < 0.001 − 0.462 < 0.001 − 0.466 < 0.001 Excess RR due to mediated interaction 0.127 < 0.001 0.130 < 0.001 0.139 < 0.001 0.145 < 0.001 0.156 < 0.001 0.151 < 0.001 0.157 < 0.001 Excess RR due to pure indirect effect − 0.261 < 0.001 − 0.261 < 0.001 − 0.280 < 0.001 − 0.287 < 0.001 − 0.294 < 0.001 − 0.290 < 0.001 − 0.297 < 0.001 Mediator at 1 Total excess RR − 0.570 < 0.001 − 0.581 < 0.001 − 0.586 < 0.001 − 0.606 < 0.001 − 0.623 < 0.001 − 0.616 < 0.001 − 0.625 < 0.001 Excess RR due to controlled direct effect − 0.590 < 0.001 − 0.600 < 0.001 − 0.600 < 0.001 − 0.615 < 0.001 − 0.639 < 0.001 − 0.620 < 0.001 − 0.625 < 0.001 Excess RR due to reference interaction 0.154 < 0.001 0.150 < 0.001 0.155 < 0.001 0.152 < 0.001 0.154 < 0.001 0.144 < 0.001 0.140 < 0.001 Excess RR due to mediated interaction 0.127 < 0.001 0.130 < 0.001 0.139 < 0.001 0.145 < 0.001 0.156 < 0.001 0.151 < 0.001 0.157 < 0.001 Excess RR due to pure indirect effect − 0.261  < 0.001 − 0.261  < 0.001 − 0.280  < 0.001 − 0.287  < 0.001 − 0.294  < 0.001 − 0.290  < 0.001 − 0.297  < 0.001 RR relative risk, COC continuity of care, DDI drug–drug interaction. aβ: model estimates. bLow COC group is the reference group. Figure 1 presents the policy-relevant proportions, setting the mediator at neither polypharmacy nor excessive polypharmacy. The proportion mediated demonstrated the importance of the mediation pathway to explain the exposure’s effect on the outcome43. The proportion eliminated captured the portion of the exposure’s effect on the outcome that would be eliminated through a policy that intervened and fixed the mediator to a certain level43. The proportion of interaction is the sum of reference interaction and mediated interaction, which showed how exposure and mediator affect the outcome variable together. We found that the proportion mediated was greater when excessive polypharmacy was used as a mediator than when polypharmacy was used. Not much change was observed in the proportion mediated during the study period. The proportion eliminated indicated that in the absence of polypharmacy, there would be no direct protection from COC, which can be harmful. However, controlling excessive polypharmacy can be benefited through enhanced COC. The proportion of interaction was greater using polypharmacy as mediator, suggesting that the interaction between COC and polypharmacy played a more essential part in preventing DDI than excessive polypharmacy.Figure 1 Policy-relevant proportions using different mediators. This figure demonstrated three policy-relevant proportions of the effect of COC on the DDI events: the proportion eliminated (INTref + INTmed + PIE/TE), the proportion mediated (INTmed + PIE/TE) and the proportion of interaction (INTref + INTmed/TE). With different mediators, the constitute of the effect of COC on DDI events differs. The proportion mediated was higher in the model with excessive polypharmacy (≥ 10 medications concurrently used) than the one with polypharmacy (≥ 5 medications concurrently used); while the proportion of interaction was greater treating polypharmacy as mediator instead. The proportion eliminated declined during the study period using excessive polypharmacy as a mediator. We stratified our analysis by sex and age and compared how the effect differed among the groups. In the analysis for sex, we found that in both models, the proportion of interaction effect was higher in women than in men, while the proportion mediated showed an opposite effect. To analyze the effect by age, we divided the study population into 3 groups: those aged 65 to 74, 75 to 84, and ≥ 85 years. We found a difference in the proportion mediated for different mediators and age groups. The proportion mediated was the highest in the group aged ≥ 85 years using excessive polypharmacy as the mediator, and it was the highest in the group aged 65 to 74 years using polypharmacy (instead of excessive polypharmacy) as the mediator. The differences between the proportions of interaction were nonsignificant in different age groups. Sensitivity analysis To test the robustness of our result, we conducted a sensitivity analysis by altering the definition of the exposure and the outcome, as described in the Methods section. Our results remained robust for all these tests, that is, higher COC was related to less DDI events, and the relationship was partially mediated by polypharmacy (tables for subgroup and sensitivity analyses are not shown; results are available upon request). We redefined polypharmacy and excessive polypharmacy as the persistent status that lasted for the entire year, and the results of the proportion eliminated and proportion mediated were different. The proportion mediated was much lower than that suggested by the original definition in both polypharmacy and excessive polypharmacy. The proportion mediated was lower in persistent excessive polypharmacy than in excessive polypharmacy. However, there was only a slight difference between persistent polypharmacy and polypharmacy. Discussion In this study, we investigated whether polypharmacy mediates the effect of COC on DDI in older patients. We found significant partial mediation effect of polypharmacy, and this mediation effect was higher when excessive polypharmacy was defined as the mediator. Our study provides evidence for maximizing the effect of COC on reducing DDI by intervening on polypharmacy. Through prescription of appropriate medication, inappropriate polypharmacy can be avoided44. Our results suggest that multiple drug use is common in the elderly population, and this problem is worse in Taiwan than in other countries and regions11,45,46. The high prevalence of polypharmacy in Taiwan is potentially attributable to high physician visit frequency. Our results demonstrated that the average number of physician visits was 30 and that approximately two-thirds of the study population had > 17 physician visits in 1 year. Because of the lack of a proper referral system in Taiwan, patients may self-refer to physicians, leading to low COC5. We found a negative association between COC and polypharmacy; that is, high COC prevented polypharmacy in patients. In addition, polypharmacy and DDI were positively correlated, confirming that polypharmacy is a risk factor for DDI. In the mediation analysis, we found that enhancing COC has protective effects for DDI events. However, when decomposing the total effect, polypharmacy and excessive polypharmacy exerted effects of different magnitudes. One study that analyzed patients with hypertension among Medicare beneficiaries found that greater COC causes elderly people to purchase more drugs for hypertension treatment47. This provides evidence for the possible existence of a mediation effect of polypharmacy. In our study, when polypharmacy and excessive polypharmacy were eliminated, the effect of COC on DDI events occurred primarily because of the interaction between COC and polypharmacy status. Although polypharmacy or excessive polypharmacy was presented, the protective effect could be attributed to the direct effect of COC. Furthermore, the level of interaction and mediation effect between polypharmacy and excessive polypharmacy was different. The proportion of pure mediation through excessive polypharmacy was higher than that through polypharmacy. The interaction between COC and excessive polypharmacy was greater than that between COC and polypharmacy. Thus, identifying the polypharmacy status of a patient may be essential. For patients having polypharmacy, enhancing their COC could be sufficient because direct effect constitutes a large portion of the effect. However, for excessive polypharmacy patients, enhancing their COC might not be enough. If physicians are unaware of the possible DDI incidence, they might prescribe more medications for symptom control, and this explains the positive effect of the COC-excessive polypharmacy interaction on DDI. Measures other than enhancing COC must be considered to manage the DDI of patients with excessive polypharmacy. In the subgroup analysis, we selected sex and age as the criteria for grouping because members of different sex have been found to demonstrate different health care-seeking behaviors40–42. Our study revealed that the components of COC’s effect on DDI events were different for men and women. Mediation is the effect whereby exposure affects the outcome through the mediator; interaction describes how the effect of the exposure on the outcome is modified by the mediator. Women consult physicians more than men do41, which is a potential explanation for our baseline description of lower COC among women than men. Other studies have revealed that women are more likely to use multiple drugs than men48–50; and a more positive attitude toward medications in women may lead to greater willingness to use drugs42. As a result, the enhancement of COC might exert a greater effect on medication management among women, further reducing the frequency of DDI events in women. In the analysis of the effect of COC on DDI events for different age groups, the effect components had a higher difference in the models that used excessive polypharmacy as the mediator. The proportion eliminated of the group aged ≥ 85 years was lower than that of other age groups. Assumption for the change could thus be made that for this age group, the treatment might be more conservative and regular14. Hence, visiting a regular physician familiar with the patient in addition to the drug history might be helpful. Managing the number of medications could be an effective way to enhance the positive effect of COC on reducing DDI. Recent studies and policies have primarily emphasized on managing potential inappropriate medications. Many guidelines such as Beer’s criteria and STOPP/START criteria were developed to reduce the number of medications administered by reviewing the medications and evaluating the appropriateness51,52. Nevertheless, appropriate medications might impose other risks on patients, such as DDIs. To manage this problem, other inventions dealing with DDI-related polypharmacy should be considered. Studies have shown that pharmacists may be crucial in reducing the polypharmacy that imposes risk on patients27,53. Regular reconciliation of medications can reduce the number of drugs and lower the burden of polypharmacy. Establishing monitoring plans for DDIs can ensure patient safety. Nowadays, electronic systems have been widely adopted in medical facilities. NHI PharmaCloud system, established in 2013, served as an effective tool for medication evaluation because it is connected to the NHI claim system, which allows for the sharing of prescription information54. Despite the platform’s ability to improve the health care quality, a survey conducted among physicians indicated resistance to use the system55. Improving the information quality also enhances people’s intention to use the tool. Setting up reminders for reviewing polypharmacy status may be helpful in alerting the medical staff to evaluate the medication usage and increase the awareness of potential DDIs. This study had limitations that should be considered when interpreting the results. First, we only used those patient characteristics (eg, sex and age) that were available in the claim database for the covariates. Other patient characteristics (eg, education) that may alter their physician-seeking and medication-using behaviors were not adjusted in this study46. Second, the claims data provided no information on medication adherence and only allowed us to identify the prescribed DDI combinations. However, given that some DDIs can be fatal, administering these combinations of DDI drugs on patients should be a public health concern, regardless of the levels of medication adherence. Finally, our results may not be generalized to health care systems that mainly adopt private insurances or sound referral systems. Different health care systems can produce different COC with varying COC effects. Our results provided useful information for countries with NHI38. We only involved some variables that had been well-studied in the COC study. However, other factors can influence the effect of COC on DDI. However, we believed that the effect of these factors are minimal. Both DDI and polypharmacy are common phenomena for the elderly population, and the magnitude of polypharmacy influences the quality of care. In this study, we found that polypharmacy was a partial mediator in the relationship between COC and DDI, and the effect of mediation differed with different magnitudes of polypharmacy. This provided the empirical proof supporting the claim that polypharmacy should be addressed in patient-caring53. Moreover, different interventions should be developed categorized by age, gender, and polypharmacy status. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Acknowledgements This study was supported by Ministry of Science and Technology, Taiwan (Grant number: 109-2314-B-010-049-MY2). Grant recipient: Christy Pu. Author contributions Y.A.W. was responsible for data analysis. Y.A.W., C.Y.D., and C.P. wrote the manuscript, C Pu supervised the study and finalized the manuscript. Competing interests The authors declare no competing interests. ==== Refs References 1. Jazbar J Locatelli I Horvat N Kos M Clinically relevant potential drug–drug interactions among outpatients: a nationwide database study Res. Soc. Adm. Pharm. 2018 14 572 580 10.1016/j.sapharm.2017.07.004 2. Obreli-Neto PR Adverse drug reactions caused by drug–drug interactions in elderly outpatients: a prospective cohort study Eur. J. Clin. Pharmacol. 2012 68 1667 1676 10.1007/s00228-012-1309-3 22644345 3. Juurlink DN Mamdani M Kopp A Laupacis A Redelmeier DA Drug–drug interactions among elderly patients hospitalized for drug toxicity JAMA 2003 289 1652 1658 10.1001/jama.289.13.1652 12672733 4. Haggerty JL Continuity of care: a multidisciplinary review BMJ 2003 327 1219 1221 10.1136/bmj.327.7425.1219 14630762 5. Chu H-Y Chen C-C Cheng S-H Continuity of care, potentially inappropriate medication, and health care outcomes among the elderly: evidence from a longitudinal analysis in Taiwan Med. Care 2012 50 1002 1009 10.1097/MLR.0b013e31826c870f 23047791 6. Chen C-C Tseng C-H Cheng S-H Continuity of care, medication adherence, and health care outcomes among patients with newly diagnosed type 2 diabetes: a longitudinal analysis Med. Care 2013 51 231 237 10.1097/MLR.0b013e31827da5b9 23269110 7. Romero-Ventosa EY Multidisciplinary teams involved: detection of drug-related problems through continuity of care Farm. Hosp. 2016 40 529 543 27894227 8. Guo J-Y Chou Y-J Pu C Effect of continuity of care on drug–drug interactions Med. Care 2017 55 744 751 10.1097/MLR.0000000000000758 28650921 9. Marengoni A Onder G Guidelines, polypharmacy, and drug–drug interactions in patients with multimorbidity BMJ 2015 11 350 10. Tamblyn RM McLeod PJ Abrahamowicz M Laprise R Do too many cooks spoil the broth? Multiple physician involvement in medical management of elderly patients and potentially inappropriate drug combinations CMAJ Can. Med. Assoc. J. 1996 154 1177 8612253 11. Slabaugh SL Maio V Templin M Abouzaid S Prevalence and risk of polypharmacy among the elderly in an outpatient setting Drugs Aging 2010 27 1019 1028 10.2165/11584990-000000000-00000 21087071 12. Cheng S-H Chen C-C Effects of continuity of care on medication duplication among the elderly Med. Care 2014 52 149 156 10.1097/MLR.0000000000000042 24309666 13. Tsuji-Hayashi Y Fukuhara S Green J Kurokawa K Use of prescribed drugs among older people in Japan: association with not having a regular physician J. Am. Geriatr. Soc. 1999 47 1425 1429 10.1111/j.1532-5415.1999.tb01561.x 10591236 14. Lu W-H Wen Y-W Chen L-K Hsiao F-Y Effect of polypharmacy, potentially inappropriate medications and anticholinergic burden on clinical outcomes: a retrospective cohort study Can. Med. Assoc. J. 2015 187 E130 E137 10.1503/cmaj.141219 25646290 15. Christakis DA Mell L Koepsell TD Zimmerman FJ Connell FA Association of lower continuity of care with greater risk of emergency department use and hospitalization in children Pediatrics 2001 107 524 529 10.1542/peds.107.3.524 11230593 16. Smedby Ö Eklund G Eriksson EA Smedby B Measures of continuity of care: a register-based correlation study Med. Care 1986 24 511 518 10.1097/00005650-198606000-00005 3713289 17. Pollack CE Measuring care continuity: a comparison of claims-based methods Med. Care 2016 54 e30 10.1097/MLR.0000000000000018 24309664 18. Bice TW Boxerman SB A quantitative measure of continuity of care Med. Care 1977 15 347 349 10.1097/00005650-197704000-00010 859364 19. Masnoon N Shakib S Kalisch-Ellett L Caughey GE What is polypharmacy? A systematic review of definitions BMC Geriatr. 2017 17 230 10.1186/s12877-017-0621-2 29017448 20. Chan D-C Hao Y-T Wu S-C Polypharmacy among disabled taiwanese elderly Drugs Aging 2009 26 345 354 10.2165/00002512-200926040-00005 19476401 21. Doan J Zakrzewski-Jakubiak H Roy J Turgeon J Tannenbaum C Prevalence and risk of potential cytochrome p450–mediated drug–drug interactions in older hospitalized patients with polypharmacy Ann. Pharmacother. 2013 47 324 332 10.1345/aph.1R621 23482734 22. Wang Y-J Is excessive polypharmacy a transient or persistent phenomenon? A nationwide cohort study in Taiwan Front. Pharmacol. 2018 9 120 10.3389/fphar.2018.00120 29515446 23. Marusic S Actual drug–drug interactions in elderly patients discharged from internal medicine clinic: a prospective observational study Eur. J. Clin. Pharmacol. 2013 69 1717 1724 10.1007/s00228-013-1531-7 23739998 24. Nobili A Potentially severe drug interactions in elderly outpatients: results of an observational study of an administrative prescription database J. Clin. Pharm. Ther. 2009 34 377 386 10.1111/j.1365-2710.2009.01021.x 19583670 25. Johnell K Klarin I The relationship between number of drugs and potential drug–drug interactions in the elderly Drug Saf. 2007 30 911 918 10.2165/00002018-200730100-00009 17867728 26. Malone DC Identification of serious drug–drug interactions: results of the partnership to prevent drug–drug interactions J. Am. Pharm. Assoc. 2004 44 142 151 10.1331/154434504773062591 27. Cornu P Steurbaut S Šoštarić S Mrhar A Dupont AG Performance of a clinical decision support system and of clinical pharmacists in preventing drug–drug interactions on a geriatric ward Int. J. Clin. Pharm. 2014 36 519 525 10.1007/s11096-014-9925-x 24566821 28. Phansalkar S High-priority drug–drug interactions for use in electronic health records J. Am. Med. Inform. Assoc. 2012 19 735 743 10.1136/amiajnl-2011-000612 22539083 29. McEvoy DS Variation in high-priority drug–drug interaction alerts across institutions and electronic health records J. Am. Med. Inform. Assoc. 2017 24 331 338 10.1093/jamia/ocw114 27570216 30. Cornu P High-priority and low-priority drug–drug interactions in different international electronic health record systems: a comparative study Int. J. Med. Inform. 2018 111 165 171 10.1016/j.ijmedinf.2017.12.027 29425628 31. VanderWeele T Explanation in causal inference: developments in mediation and interaction Int. J. Epidemiol. 2016 45 1904 1908 27864406 32. Baron RM Kenny DA The moderator–mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations J. Pers. Soc. Psychol. 1986 51 1173 10.1037/0022-3514.51.6.1173 3806354 33. Judd CM Kenny DA Process analysis: estimating mediation in treatment evaluations Eval. Rev. 1981 5 602 619 10.1177/0193841X8100500502 34. Jiang Z VanderWeele TJ When is the difference method conservative for assessing mediation? Am. J. Epidemiol. 2015 182 105 108 10.1093/aje/kwv059 25944885 35. Valeri L VanderWeele TJ Mediation analysis allowing for exposure–mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros Psychol. Methods 2013 18 137 10.1037/a0031034 23379553 36. VanderWeele TJ A unification of mediation and interaction: a four-way decomposition Epidemiology 2014 25 749 10.1097/EDE.0000000000000121 25000145 37. Discacciati A Bellavia A Lee JJ Mazumdar M Valeri L Med4way: a Stata command to investigate mediating and interactive mechanisms using the four-way effect decomposition Int. J. Epidemiol. 2018 48 15 20 10.1093/ije/dyy236 38. Shin DW Impact of continuity of care on mortality and health care costs: a nationwide cohort study in Korea Ann. Fam. Med. 2014 12 534 541 10.1370/afm.1685 25384815 39. Chan DC Hao YT Wu SC Characteristics of outpatient prescriptions for frail Taiwanese elders with long-term care needs Pharmacoepidemiol. Drug Saf. 2009 18 327 334 10.1002/pds.1712 19180586 40. Hong JS Kang HC Kim J Continuity of care for elderly patients with diabetes mellitus, hypertension, asthma, and chronic obstructive pulmonary disease in Korea J. Korean Med. Sci. 2010 25 1259 1271 10.3346/jkms.2010.25.9.1259 20808667 41. Bertakis KD The influence of gender on the doctor–patient interaction Patient Educ. Couns. 2009 76 356 360 10.1016/j.pec.2009.07.022 19647968 42. Hofer-Dückelmann C Gender and polypharmacotherapy in the elderly: a clinical challenge Handb. Exp. Pharmacol. 2012 214 169 182 10.1007/978-3-642-30726-3_9 43. VanderWeele TJ Policy-relevant proportions for direct effects Epidemiology (Cambridge, Mass.) 2013 24 175 10.1097/EDE.0b013e3182781410 44. Scott IA Reducing inappropriate polypharmacy: the process of deprescribing JAMA Intern. Med. 2015 175 827 834 10.1001/jamainternmed.2015.0324 25798731 45. Wastesson JW Morin L Laroche ML Johnell K How chronic is polypharmacy in old age? A longitudinal nationwide cohort study J. Am. Geriatr. Soc. 2019 67 455 462 10.1111/jgs.15717 30575952 46. Midão L Giardini A Menditto E Kardas P Costa E Polypharmacy prevalence among older adults based on the survey of health, ageing and retirement in Europe Arch. Gerontol. Geriatr. 2018 78 213 220 10.1016/j.archger.2018.06.018 30015057 47. Robles S Anderson GF Continuity of care and its effect on prescription drug use among Medicare beneficiaries with hypertension Med. Care 2011 49 516 521 10.1097/MLR.0b013e31820fb10c 21430582 48. Manteuffel M Influence of patient sex and gender on medication use, adherence, and prescribing alignment with guidelines J. Women's Health 2014 23 112 119 10.1089/jwh.2012.3972 49. Venturini CD Gender differences, polypharmacy, and potential pharmacological interactions in the elderly Clinics 2011 66 11 1867 1872 22086515 50. Schuler J Polypharmacy and inappropriate prescribing in elderly internal-medicine patients in Austria Wien. Klin. Wochenschr. 2008 120 733 10.1007/s00508-008-1089-z 19122984 51. Panel AGSBCUE American Geriatrics Society 2019 updated AGS Beers Criteria® for potentially inappropriate medication use in older adults J. Am. Geriatr. Soc. 2019 67 674 694 10.1111/jgs.15767 30693946 52. O'Mahony D STOPP/START criteria for potentially inappropriate prescribing in older people: version 2 Age Ageing 2015 44 213 218 10.1093/ageing/afu145 25324330 53. Chumney EC Robinson LC The effects of pharmacist interventions on patients with polypharmacy Pharm. Pract. 2006 4 103 54. Huang S-K NHI-PharmaCloud in Taiwan—a preliminary evaluation using the RE-AIM framework and lessons learned Int. J. Med. Inform. 2015 84 817 825 10.1016/j.ijmedinf.2015.06.001 26113462 55. Hsieh P-J Lin W-S Explaining resistance to system usage in the PharmaCloud: a view of the dual-factor model Inf. Manag. 2018 55 51 63 10.1016/j.im.2017.03.008