Polypharmacy and Unplanned Hospitalizations in Patients with Rheumatoid Arthritis

DISCUSSION

Ours is the first study, to our knowledge, to analyze the association between PP and hospitalization specifically in patients with RA. Similarly, while data on ADR as a cause of admissions are available for the general population^{11,12,23,24,25}, the data for patients with RA are missing.

It is difficult to estimate the prevalence of PP in the general population given different definitions of PP and study populations analyzed (primary/secondary care, hospital admissions, elderly population, comorbidity burden, etc.). A study using primary care data from the general population showed that almost 50% of patients aged > 20 years admitted to the hospital were prescribed at least 1 regular medication, with 25.2% receiving 1–3, 16.9% receiving 4–9, and 4.6% ≥ 10 medications²⁶. Increasing numbers of regular medications are seen with female sex, older age, greater socioeconomic deprivation, and increasing multimorbidity. Cardiometabolic conditions are the most important disease cluster^22,26. We observed that PP is common in patients with RA and is more frequently observed in women, and patients with higher disease activity, higher levels of functional impairment, and longer disease duration. None of these observations is unexpected; however, the absolute numbers of patients receiving PP is high, especially in the elderly.

Patients with higher levels of PP had a substantially greater risk of unplanned hospitalization. The risk of hospitalization was more marked in patients receiving > 10 medications. The observation that the PP and hospitalization risk was nonlinear is particularly relevant if one were to attempt to use PP as an epidemiological surrogate for comorbidity. However, at a patient level, it is important to consider why the pattern is observed. It may be that the relationship between PP and comorbidity becomes stronger above a certain point, especially because we included DMARD in the total medication count.

In addition to the relationship with comorbidity, PP may be a direct causal factor in hospitalization through ADR. It has previously been shown that ADR account for between 0.5% and 6.5% of admissions^10,11,22. Our estimate of definite or probable ADR-related admissions within an RA cohort was 6.9%. The low absolute numbers of definite or probable ADR precluded detailed analysis; however, it appeared that ADR were linked to specific classes of high-risk drugs (e.g., anticoagulants), whose effects may be potentiated by specific comorbidities and drug interactions that increase bleeding risk, rather than absolute numbers of medication.

Some ADR can be avoidable (hazardous prescribing), but we observed very few such events. In contrast, many ADR occur in the setting of appropriate prescribing. In the context of RA, most clinicians now accept that the infectious risks of DMARD are far outweighed by the beneficial effect upon RA, with the knowledge that untreated RA is a far more hazardous state.

Our analyses, with specific consideration for RA drugs within the context of PP, were reassuring, with no demonstrated apparent association between more intensive DMARD strategies, including use of biologics, and hospitalization risk. There was, however, a 2-fold higher rate of hospitalization among corticosteroids users. These could have been prescribed either for treatment of RA or another comorbidity, increasing the risk of hospitalization; however, owing to lack of information on comorbidities in our study we are unable to draw conclusions on an indication. Although DMARD contribute to PP, it may be that an adequate control of disease activity is protective against admission. An alternative explanation is that there may be channeling bias, with healthier patients more likely to be prescribed combination DMARD or biologic treatment strategies, while clinicians adopt more cautious approaches (perhaps favoring steroids) for patients with complex background comorbidity.

Drug-drug interactions were documented in 82% of admissions, but only 11% were considered of clinical significance. As expected, patients in higher PP strata were more exposed to drug-drug interactions compared to patients in the lowest strata. However, it is important to acknowledge that simple assessments of drug-drug interaction may dramatically overestimate risk of clinically relevant problems²⁷.

PP may be a useful proxy tool (unlike the calculation of indices of comorbidity) to adjust for confounding by comorbidity in epidemiologic analyses and identify patients at high risk of hospitalizations for targeted risk management. Comorbidity indices have been used to measure and weigh the overall burden of comorbidities and have been used in predicting mortality²⁸, but not in predicting acute admissions. Different methods have been used to predict risk of emergency admissions, which took into account demographic, lifestyle, laboratory, and clinical variables, and chronic disease^23,29. These tools appear complicated for stratification of acute admissions in daily practice, and routinely collected PP data may act as a useful surrogate. Whether this is the case could be investigated in further studies.

A key strength of this research is the use of real-world data — electronic medical records (EMR) and emphasis on collecting defined core data generated in day-to-day situations rather than typically selective controlled trials. The growing use of EMR and emphasis on collecting defined core data in specific diseases such as RA makes it easier to use real-world data sources in research. Using routinely collected data to evaluate the effect of PP has both strengths and limitations. The growing use of EMR and routine data sources may help overcome issues of generalizability that limit trials.

Limitations of our study include misclassification biases, unmeasured confounding, missing data, and censorship. Clinicians may be less thorough in recording medication in routine care than in formal clinical studies, resulting in reduced data quality or missing data. Because we relied upon the secondary care record of patient-prescribed medication, it is likely that we were unaware of some medications patients were receiving (including over-the-counter medications), and our measurement of PP is likely an underestimate. However, for all hospitalizations, a pharmacist undertakes medicine reconciliation directly with primary caregivers and, therefore, we were able to compare an actual number of prescribed medications with our own record of PP at baseline. Among patients admitted with an infection, the mean number of medications recorded in our baseline record was 7.0 (SD 4.0), compared to a mean number of medications of 8.9 (SD 4.3) at the time of admission. While some of this increase may genuinely represent additional medications prescribed between baseline and the date of admission, it also provides some estimate of the extent of missing information on baseline medication.

Regarding misclassification of exposure, recall bias was avoided by using electronically recorded prescription data, but precise information on dispensing and adherence was unavailable. Indeed, it was inaccuracy in comorbidity coding that initiated the authors’ interest in PP as a measure. It is notable that when we compared the medicine reconciliation data from pharmacists at the time of admission, the extent of medication misclassification was reassuringly low. However, a number of nuances, such as transient episodic use of medications, may have been missed. A further limitation is that it is likely that some patients had admissions at other hospitals. We hypothesize that additional admission to other inner-city Trust hospitals may have contributed to the higher number of drug-related admissions. Therefore, our estimates of hospitalization rate can be considered conservative because we will have underestimated the true rate. However, there is no reason to think that this bias would have affected patients at different levels of PP differentially. We attempted to reduce confounding by design and analysis (multivariable adjustments on age and sex); however, residual confounding likely remains (e.g., socioeconomic and other factors). The complexity of medications in some cases and the lack of full medical records made the judgment of appropriate or inappropriate prescribing impossible.

Our study found PP to be associated with increased risk of acute hospitalizations, particularly for those taking > 10 medications. There are 2 likely drivers for these effects: PP as a contributing cause of an increased drug-related toxicity or PP as an indicator of greater comorbidity burden. It is unclear in what manner PP influences adverse outcomes and our data suggest the risk may be nonlinear. However, PP should undoubtedly be closely monitored to minimize potentially inappropriate combinations of prescribed medications. These observations also support the hypothesis that PP may be a useful clinical tool: a simple, novel, and readily measurable predictor of clinical outcomes. Undoubtedly, more research and further validation studies need to be done before firm conclusions can be drawn. However, patients exposed to higher levels of PP represent a population of particular relevance to the modern NHS: these patients are typically excluded from clinical trials and, therefore, data regarding drug efficacy in the setting of PP are lacking²⁷.