Introduction

Since COVID-19 first emerged in Wuhan, China in December 2019, it has caused a global pandemic with a staggering 418 million infections and 5.9 million fatalities reported as of 18 February 2022.1 2 It is well known that older adults are at a greater risk of mortality and severe disease than younger people.3–5 The fact that older patients have several comorbid conditions such as hypertension and diabetes along with a weakened immune system must have put them at greater risk.6 7 In the epidemiology of COVID-19, men, low-income people and individuals with underlying comorbidity have a greater risk of morbidity and mortality.8–11

The prevalence of potentially inappropriate medication (PIM) use in older patients is alarming and varies between 18.5% and 79% depending on the population studied.12–17 PIM use has been linked to increased risk of adverse drug reactions, which compromises therapeutic benefits.18–20 These medications are also associated with negative health outcomes such as increased mortality and morbidity.21–26

PIMs are recognised for their capacity to worsen disease prognosis due to their pharmacological effects. Specifically, benzodiazepine-based PIMs have been linked to an elevated risk of cognitive impairment and delirium, attributed to their actions on the central nervous system.27 28 Similarly, anticholinergic PIMs are associated with an increased risk of dementia, as they block the neurotransmitter acetylcholine.20 When older adults are exposed to PIMs, they are undeniably vulnerable to these risks.

These individuals, if they contract COVID-19, could face more precarious health outcomes due to additional COVID-19-related risks. Specifically, benzodiazepine-based PIMs have the potential to reduce respiratory muscle strength and endurance,29 which could exacerbate the respiratory challenges posed by COVID-19. Moreover, PIMs involving proton pump inhibitors are suspected to heighten the risk of ventilator-related adverse events, likely due to bacterial overgrowth resulting from the inhibition of gastric acid secretion.30–32 Additionally, PIMs comprising non-steroidal anti-inflammatory drugs might impair the immune system’s ability to fight infections by suppressing cytokine induction.33 Given these concerns, it is crucial to determine whether older patients with COVID-19, when exposed to PIMs, face an increased risk of experiencing a poor prognosis

This study aimed to investigate the temporal association between exposure to PIM and poor prognosis of COVID-19 in older adults, taking into account risk factors, using the National Health Insurance Service (NHIS)-COVID-19 cohort database. The results of this study will highlight the importance of considering PIM exposure in older COVID-19 patients and could inform guidelines for their pharmacological treatment, as current studies on this topic are limited.34 35

MethodsStudy design

We conducted a retrospective cohort study using data from the NHIS-COVID-19 database, which links health insurance claims to the Korea Disease Control and Prevention Agency (KDCPA) epidemic registry. The registry meticulously records healthcare resource utilisation, including inpatient and outpatient services for patients infected from 20 January 2020 to 4 June 2020, with data documented up to 31 July 2020. The national health insurance claims data contain information on insurance claims made for covered healthcare utilisations before infection, but do not contain lab values and disease severities.

In the study, older adults who were exposed to PIM during 30 days prior to a positive test for COVID-19 were identified along with those who were not exposed. The 30 days exposure time window was chosen based on similar studies done on the PIM’s risk.36–38 The cohorts were matched based on propensity scores and were followed for events of critical care services and deaths until release from isolation (figure 1). This retrospective cohort study was approved by the Seoul National University Institutional Review Board (IRB No. E2103/004-004).

Figure 1

Nationwide cohort study design. NHIS, National Health Insurance Service; PIM, potentially inappropriate medication.

Study population

Individuals aged 60 and above who tested positive for SARS-CoV-2 from 20 January 2020 to 4 June 2020 were included in this study (n=2217).

Patient and public involvement

It was not appropriate or possible to involve patients or the public in the design, or conduct, or reporting, or dissemination plans of our research

Exposure to PIM

The American Geriatrics Society has identified and published various PIMs in the 2019 Beers criteria.39 The published list of the medications that are potentially inappropriate in most older adults classifies PIMs into nine drug categories (anticholinergics, antithrombotics, anti-infectives, cardiovascular, central nervous system, endocrine, gastrointestinal, pain-medications and genitourinary).39 In the list, Korea’s National Drug Formulary recognises 88 active pharmaceutical ingredients (APIs) for drug insurance claims. However, this number was reduced to 68 out of 8 drug categories when their use for each indication was evaluated (online supplemental table 1). For instance, antipsychotics were excluded from the PIM list because they are mainly prescribed for schizophrenia or bipolar disorders in Korea.40 After finalising the PIM list, exposure to PIM was determined based on any record of prescription in the PIM list within 30 days prior to the infection.

The PIM list was exhaustively identified, and the prescription records for each patient that tested positive for COVID-19 based on the NHIS drug codes during the exposure time window were examined.

Disease outcomes of COVID-19

Each study subject was followed from the time of confirmed positive test result to the time of release from isolation for observation of events requiring critical care or death. NHIS codes were used to identify utilisation of critical care services such as intensive care unit, artificial ventilation, extracorporeal membrane oxygenation and cardiopulmonary resuscitation41 42 (online supplemental table 2). Deaths were identified from KDCPA epidemic registry during the infection period.

Propensity score matching

The cohorts (PIM-exposed and PIM-non-exposed) were 1:1 matched based on greedy propensity score matching with calliper width of 0.1 to balance baseline characteristics.43 44 The propensity scores were constructed from variables such as age, sex, region of residence, economic status, Charlson Comorbidity Index (0, 1, 2, ≥3)45 46 and concomitant diseases. Charlson Comorbidity Index was computed based on the ICD-10 (International Classification of Diseases, 10th revision) codes linked to the conditions for which each patient had made healthcare visits 1 year prior to testing positive for COVID-19 (online supplemental table 3).47–49 The concomitant diseases identified were hypertension, diabetes, hyperlipidaemia, cardiovascular disease, peripheral vascular disease, cerebrovascular disease, chronic obstructive pulmonary disease, asthma, liver disease, renal disease, mental disorder, musculoskeletal disorder, parkinsonism, dementia, gastrointestinal disorder and cancer (online supplemental table 4).48–50 The quality of balancing between two cohorts was assessed from computing the absolute standardised mean difference (ASMD). If ASMD is less than 0.1, the two cohorts are well balanced after propensity score matching.43 44

Sensitivity and subgroup analyses

Sensitivity analysis was performed by varying the exposure time window and the propensity score matching algorithm. The exposure time window was varied from 60 to 90 days prior to the confirmed test. Propensity score matching was repeated for different matching ratios and callipers, that is, 1:2 greedy matching with a calliper of 0.1 and 1:1 greedy matching with a calliper of 0.01. Subgroup analyses were performed for stratification of age, concomitant diseases and number of PIMs. Age stratification was done for three age groups (60–69, 70–79, 80 or higher). Concomitant diseases were stratified into three groups (hypertension, diabetes mellitus and hyperlipidaemia); the selection of which was based on previous studies.42 50–52 Stratification of PIM number was done by grouping the number of PIM categories as well as the number of PIM APIs.

Statistical analysis

The baseline characteristics between the cohorts were compared for their closeness in distribution using the ASMD. For comparison of clinical outcomes between the two cohorts, t-test and chi-square were used. Three logistic regression models were run to estimate the OR of risk of mortality and critical care use between the cohorts along with their 95% CI. Logistic regression model 1 included sociodemographic, comorbidity and concomitant diseases as covariables, in addition to the independent variable of the PIM exposure indicator without propensity score matching. Models 2 and 3 logistic regressions were run for propensity score-matched pairs. However, model 3 logistic regression had covariates such as pharmacological treatment and hospitalisation. Pharmacological treatments were those mentioned in the COVID-19 treatment guideline53 (online supplemental table 5). These covariates could not be used for propensity score matching because they did not exist prior to the identification of PIM exposure. The ORs were converted to relative risks (RRs) to align with a cohort study design.54 55 All data were analysed using SAS Enterprise Guide software (V.7.2).

Patient and public involvement

This study used the NHIS-COVID-19 database, which links health insurance claims to the KDCPA epidemic registry. Therefore, there was no patient or public involvement in this study.

ResultsCohort description

The cohort comprised of 2217 patients aged 60 years and above who tested positive for SARS-CoV-2 during the 6 months following its outbreak in Korea (20 January 2020 and 4 June 2020). Of those, 604 were exposed to PIM (27.2%) during the 30 days prior to their infection with the virus (table 1). Majority of patients who had PIM exposure resided in Daegu metropolitan area where the outbreak was first reported (65.9% for the exposed vs 68.7% for the non-exposed). However, the PIM-exposed individuals were more likely to be from the lowest income quintile than the PIM-non-exposed individuals were (36.4% vs 25.9%). The exposed were also more likely to have severe comorbidities with a Charlson Comorbidity Index score greater than 3 than the non-exposed were (58.1% vs 36.6%) (table 1, online supplemental figure 1). Additionally, the exposed individuals had higher rates of hypertension (64.7% vs 49.2%), diabetes (45.9% vs 32.4%) and hyperlipidaemia (63.1% vs 54.6%) compared with the non-exposed individuals.

Table 1

Baseline characteristics and clinical outcomes of study subjects

The number of PIMs that the exposed individuals had taken during the 30-day time window was on average 1.9 APIs for 1.5 drug categories (online supplemental figure 2). The most commonly used PIM was metoclopramide (20.7%), followed by diclofenac (19.4%), lorazepam (15.9%), diazepam (14.4%) and zolpidem (8.6%). Central nervous system drugs topped the list of most commonly used PIM drug classes (46.0%) followed by pain medication (43.9%) and gastrointestinal drug (33.8%).

The greedy matching propensity score of 1:1 ratio generated 583 pairs of PIM-exposed and PIM-non-exposed cohorts. Differences between the cohorts disappeared after the propensity score matching, except for those of gastrointestinal disorders. ASMD was lower than 0.1 for all baseline and clinical characteristics except for gastrointestinal disorders which was 0.105 (table 1).

Comparison of Healthcare utilization between cohorts

The use of healthcare services after a confirmed positive result for infection was quite different in the cohorts matched for propensity scores (table 1, online supplemental table 6). The rate of hospitalisation was lower among the PIM-exposed than among the non-exposed (82.5% vs 96.9%, p<0.001). Use of pharmacological treatments, especially lopinavir and ritonavir, for COVID-19 was also less prevalent in the exposed than in the non-exposed (69.0% vs 79.1%, p=0.001) (49.1% vs 58.3%, p=0.0015). The medication count prescribed post-infection was significantly higher among the exposed (39.8 vs 33.1, p=0.042). Furthermore, the duration of hospital stay was significantly higher among the exposed (24.5 vs 27, p=0.0257). Also, there was no significant difference in length of isolation between the cohorts (31.6 vs 30.8, p=0.4692).

Temporal association between PIM exposure and poor prognosis of COVID-19

The mortality rate was significantly higher in individuals exposed to PIM (19.7%) compared with those who were not exposed (9.8%) (table 1). The use of critical care services, particularly intensive care, was also more prevalent in the PIM-exposed individuals (10.5%) compared with the non-exposed (6.0%) (p=0.0156). However, there was no significant difference in the use of artificial ventilation, extracorporeal membrane oxygenation or cardiopulmonary resuscitation between the two groups. Logistic regression analysis confirmed that PIM-exposed individuals had a higher risk of mortality compared with the non-exposed, with the RRs ranging from 1.68 to 2.0 across the three models. Similarly, the probability of using critical care services was also found to be higher in the PIM-exposed individuals with the highest RR being 1.75 (95% CI: 1.26 to 2.39) in model 3 (table 2).

Table 2

Severe clinical outcomes associated with PIM use among older COVID-19 patients

Sensitive and subgroup analysis

The relationship between PIM exposure and risks of mortality and critical care use remained significant across various exposure windows and matching ratios (figure 2). A narrower time window had a stronger association than a wider window; the relative risk (95% CI) was 1.56 (1.13–2.10) for 60 days and 1.48 (1.08–2.01) for 90 days. The 1:2 matching ratio showed a stronger association than 1:1, with relative risks of 1.89 (1.43–2.40) and 1.62 (1.14–2.26) respectively.

Figure 2

Sensitivity analysis among older COVID-19 patients. *Relative risk was analysed in model 3. # of pts, number of patients; PIM, potentially inappropriate medication; PSM, propensity score matching.

In the age-stratified analysis, the oldest cohort (80 or higher) had the weakest temporal association between PIM exposure and mortality risk (figure 3), with a relative risk of 1.62 (1.15–2.16). In contrast, the cohorts in their 60s and 70s had stronger associations, with relative risks of 1.76 (0.74–4.02) and 1.80 (0.94–3.23) respectively. The youngest cohort had the strongest association with critical care use (RR=2.00, 95% CI: 1.12 to 3.40), followed by the 70s (RR=1.76, 95% CI: 0.92 to 3.13) and the oldest cohort (RR=1.56, 95% CI: 0.93 to 2.46) (figure 3).

Figure 3

Subgroup analysis among older COVID-19 patients. *Relative risk was analysed in model 3. #, number; API, active principal ingredient; DM, diabetes mellitus; HL, hyperlipidaemia; HTN, hypertension; PIM, potentially inappropriate medication.

In the disease-stratified analysis, each cohort with a concomitant disease was at higher risks of mortality and critical care use than the cohort without the disease. For example, the temporal association between PIM exposure and mortality risk was stronger in the cohort with a hyperlipidaemia than in the cohort without the disease (RR of mortality: 1.89 vs 1.39, RR of critical care use: 2.27 vs 1.18). Furthermore, as the number of PIMs increased in terms of APIs, the association became significantly stronger (RR (95% CI): 1.17 (0.77~1.73) for 1 API vs 2.23 (1.60~3.02) for 2 or more APIs).

Discussion

Our study revealed a significant increase in the mortality rate and utilisation of critical care services among older adults with COVID-19 who were exposed to PIM compared with those not exposed. While previous studies have explored the link between PIM and elevated mortality in older adults,23 26 none have specifically examined this association among older adults with COVID-19. In a recent study by Durmus et al, an investigation into the relationship between PIM and in-hospital mortality among older adults with COVID-19 did not identify a significant link.56 Notably, their study’s limited scope as a single-centre cross-sectional study may have restricted its generalisability. In contrast, our study relies on a propensity score-matched nationwide cohort, enhancing its representativeness.

Although older adults are more susceptible to SARS-CoV-2 infection due to their weakened immune system6 7 57 and exhibit a higher likelihood of poor COVID-19 outcomes compared with younger counterparts,34 the reasons behind their increased vulnerability when exposed to PIM remain unclear. The hypotheses previously employed to explain the link between polypharmacy and an elevated risk of mortality and hospitalisation among older adults with COVID-19 may also be applicable since polypharmacy serves as a strong indicator for PIM.58 These hypotheses include frailty associated with polypharmacy and iatrogenic risks stemming from polypharmacy. However, the frailty hypothesis can be rejected in this study because our study attempted to balance factors influencing PIM exposure via propensity scores. Its complete rejection, however, may not be possible because factors affecting frailty, such as cognitive impairment, functional status, social isolation, physical function/mobility and weight loss, were not accounted for. The iatrogenic effects of PIM appear to play a significant role in increasing the risk of mortality and critical care utilisation among older adults with COVID-19. This aligns with established evidence that PIM can lead to drug-related issues such as adverse drug reactions and drug-drug interactions,19 24 exacerbating the clinical condition of COVID-19. In the context of COVID-19, PIMs, including antidepressants and benzodiazepines, have been reported to cause drug-drug interactions.59

Our study further revealed that the impact of PIM on poor COVID-19 prognosis diminishes with age, a trend consistent with the decreasing impact of polypharmacy on mortality observed in a prior study.58 Given the iatrogenic effect, the susceptibility to adverse drug reactions and drug interactions may peak between ages 60 and 64 and subsequently decline due to a potentially weakened immune system in older age groups. Further research is required to confirm this pattern while considering the influence of polypharmacy in older age groups.

Interestingly, our study found a stronger temporal association when PIM exposure was assessed within a 30-day window compared with a 90-day window preceding infection. This observation suggests a potential effect related to the concentration of PIM. The presence of PIM within the 90-day window does not necessarily indicate prolonged exposure but may reflect a single exposure during that period. Consequently, the concentration of PIM might be weaker within the 90-day exposure window compared with the 30-day window. This aligns with findings from other studies that have reported varying results with different PIM measurement intervals.37 60 Moreover, our study indicated that patients with COVID-19 exposed to a greater number of PIMs in terms of APIs and drug categories faced a higher risk of mortality and utilisation of critical care services. This aligns with the logical expectation that iatrogenic effects are more likely when more APIs and drug categories comprise PIMs.

While our study did not consider hospitalisation as a poor prognostic marker, typically indicative of poor infection outcomes, it is essential to recognise the specific context in Korea. In Korea, treatment guidelines recommend hospitalisation for all older patients with COVID-19, regardless of disease severity.61 62 Consequently, the high hospitalisation rate among older patients (89.7%) may be driven more by capacity constraints than disease severity, similar to the utilisation of pharmacological treatment.53 63

Study limitations

Several limitations of our study warrant consideration. Notably, we did not account for frequency and dosing information concerning PIM, despite their likely impact on exposure levels. Furthermore, PIM exposure identified based on medical claims may be biased because medical claims do not necessarily mean the medications taken. A delay in diagnosis of COVID-19 could also introduce a bias to the measurement of PIM exposure. In other words, PIMs that have been initiated more than 30 days prior to the diagnosis should have been identified as an exposure to PIM had there been no delay in the diagnosis of COVID-19. Additionally, our reliance on administrative data introduces the potential for misclassification regarding critical care utilisation and deaths, although this misclassification is likely non-differential concerning exposure. Although adjustment was made for many variables, it is possible that residual confounders between the groups could still be present and that propensity score matching may not be able to balance all unmeasured confounders. They are geriatric-related variables such as frailty, functional status and mobility as well as COVID-19-related risk factors for complications such as Body Mass Index and obesity. Finally, some of the patients included in this study may have not received adequate care because of capacity restraints in hospital beds and pharmacological treatments, which could confound the study results.

Our study’s findings are specific to the period from 20 January 2020 to 4 June 2020 and may not reflect recent changes in COVID-19 treatment guidelines. Additionally, it is important to consider that current mortality rates might differ from those observed during the initial outbreak, primarily due to increased vaccination rates and the emergence of less virulent virus variants. It is worth noting that the applicability of our findings is also limited to the Beers’ list of PIMs most commonly used for older adults rather than the comprehensive list of all PIMs mentioned in the Beers’ list.

Despite these limitations, it is worth noting that this study is pioneering in its identification of the association between PIM exposure and the prognosis of COVID-19 in older adults. Furthermore, our research encompasses the entire older adult population in South Korea, as the linked database between the NHIS and the KDCPA includes all individuals who tested positive for COVID-19. As a result, this study carries significant implications for both the KDCPA and the NHIS regarding the critical importance of managing PIM.

South Korea has implemented several national insurance programmes aimed at promoting the appropriate use of medications, such as the Drug Utilisation Review systems and telephone/home-visit medication counselling.64 65 However, currently, there are no dedicated programmes in place to prevent PIMs. Preventing the use of PIMs in COVID-19-infected patients during treatment can be challenging because these medications may already be in use. To address this, a potential recommendation could be to restrict the days of supply for PIMs, especially for benzodiazepines with prolonged tapering effects. In any case, the findings of this study serve as a crucial call to action, highlighting the urgent need to explore strategies for preventing PIM exposure among older adults. By addressing this issue, we can mitigate the heightened risk of poor COVID-19 outcomes in this vulnerable population.

Conclusion

This study highlights the significant impact of PIM exposure on the prognosis of COVID-19 in older adults. Healthcare providers should be aware that older patients with COVID-19, who are already at high risk for poor outcomes, are further impacted by PIM exposure. Effective interventions are needed to address PIM exposure and improve health outcomes in this vulnerable population.