Abstract
Background and aims
The temporal relationship between potentially inappropriate medication (PIM) use and hospitalization remains uncertain. We examined whether current PIM use increases the rate of hospitalization and estimated the rate of hospitalization during exposure to individual PIMs.
Methods
A retrospective population‐based cohort study of 1 480 137 older adults was conducted using the 2003–2013 Italian Emilia‐Romagna Regional administrative healthcare database (~4.5 million residents), which includes demographic, hospital and outpatient prescription information. Each day of follow‐up was defined as exposed/unexposed to PIMs that ‘should always be avoided’, according to the Maio criteria, an Italian modified version of the Beers criteria. The study outcome was all‐cause hospitalizations. Crude PIM‐related hospitalization rates were calculated for individual PIMs. Repeated‐events Cox proportional hazards models with time‐dependent covariates estimated adjusted hazard ratios for hospitalization during PIM exposure, as defined by three versions of the Maio criteria (v2007, v2011, v2014).
Results
During >10 million person‐years of follow‐up, 54.2% of individuals used ≥1 PIM and 10.9% of all person‐time was exposed to v2014 PIMs. Among 1 604 901 hospitalizations, 15.6% occurred during v2014 PIM exposure. Crude hospitalization rates during v2014 PIM‐exposed and unexposed person‐time were 228.1 and 152.1 per 1000 person‐years, respectively. The PIM with the highest rate of hospitalization was ketorolac, while nonsteroidal anti‐inflammatory drugs had the most exposure time. The hazard of hospitalization was 16% greater (hazard ratio = 1.16; 95% confidence interval 1.14, 1.18) among patients exposed to v2014 PIMs. The v2007 and v2011 estimates were similar.
Conclusions
In this large population‐based cohort of older adults, we found a 16% increased hospitalization risk associated with PIM exposure.
Keywords: hospitalization, older adults, pharmacoepidemiology, prescribing
What is Already Known about this Subject
Several criteria exist for evaluating the appropriateness of medication use in older adults.
Earlier research has reported an increased hospitalization risk among patients initiating potentially inappropriate medications.
Prior intent‐to‐treat approaches have followed patients for short periods and have not accounted for the intermittent nature of medication use.
What this Study Adds
This study followed >1 million adults for over a decade, capturing periods of potentially inappropriate medication use as a time‐varying exposure.
The hazard of hospitalization was 16% greater during periods of potentially inappropriate medication use.
Under a theoretical scenario with zero PIM exposures, 27 444 (1.7%) fewer hospitalizations would have occurred.
Introduction
The use of medications deemed potentially inappropriate due to an unfavourable benefit‐to‐risk profile represents an endemic problem affecting older adults globally 1, 2, 3, 4. Inappropriate prescribing poses a particular risk for the older adult population owing to various factors, including cognitive and physiological changes associated with the ageing process. The increased exposure to potentially inappropriate medications (PIMs) among the elderly is inextricably linked to the larger burden of multimorbidity and polypharmacy in this population 5, 6. Interactions between a complex medication regimen, physiological changes due to ageing (e.g. altered drug metabolism) 7 and multiple chronic diseases coalesce to inflate the risk of harm for a patient exposed to PIMs.
Sets of explicit guidance have been developed to evaluate medication use further and address the challenge of selecting appropriate medications for the older adult population. Practice‐based guidance documents, including the Beers criteria and the Screening Tool of Older Persons’ Potentially Inappropriate Prescriptions (STOPP) criteria, are available to guide prescribing 8, 9. These criteria classify medications based upon the estimated risk of clinical harm vs. clinical benefit 8, with PIMs defined by an estimate of net harm when the medication is used for the population under consideration. These criteria can be used as clinical guidelines to evaluate the prevalence of PIM exposure, develop interventions and evaluate postintervention trends 1, 8, 9, 10, 11. Ultimately, the main purpose of these criteria is to improve clinical outcomes; however, the links between PIM exposure and prominent health outcomes of interest to patients and health system stakeholders remain to be fully elucidated.
Several recent analyses of administrative claims have found patients with PIM exposure to have an increased risk of hospitalization compared with unexposed patients 3, 12, 13, 14, 15, while other hospital‐based studies have reported rates of PIM‐related hospitalization to be relatively low 16, 17, 18. Throughout this body of literature, the temporal association between PIM exposure and outcome has been heterogeneously defined, and a full accounting of exposed person‐time for the purpose of incidence rate estimation and quantification of the relative velocity of adverse outcomes among current PIM users and non‐users has been largely unaddressed. Therefore, in the absence of a time‐dependent accounting of exposure and confounding factors, a level of ambiguity exists about whether, and the magnitude to which, PIMs truly increase the risk of patient harm during the period of exposure.
Leveraging a large, comprehensive, longitudinal database in the Emilia‐Romagna region of Italy, we sought to evaluate the relationship between PIM exposure and hospitalization. Specifically, our aim was to determine whether current PIM exposure increases the rate of hospitalization, while defining PIM exposure using an Italian‐modified version of the Beers criteria – namely, the Maio criteria 1, 10, 11, 19. Additionally, we estimated the rate of hospitalization during exposure to each individual PIM.
Methods
Setting
Italy has a single‐payer healthcare system, whereby healthcare is delivered by the 21 regional governments through a network of geographically defined local health authorities. Each patient is registered with a general practitioner or a paediatrician within a specific local health authority 20, 21.
Study design and data source
We conducted a retrospective cohort study using the Emilia‐Romagna regional database, which contains de‐identified, fully linked, longitudinal administrative claims data for around 4.5 million inhabitants. The Emilia‐Romagna database includes demographics, hospital data and outpatient prescription information. The pharmacy records capture all prescriptions reimbursed by the Italian National Formulary and filled in an outpatient setting, which are identified using the Anatomical Therapeutic Chemical (ATC) classification system 22 and the coding system of the Italian Ministry of Health, which contains a unique numerical code for each medication available in Italy.
Study population
The study period encompassed 1 January 2003 until 31 December 2013, with baseline information collected beginning 1 January 2002. Patients aged 65 years and older who resided in the region for the entirety of 2002 entered the study cohort at the start of follow‐up. The cohort remained open for the duration of the study period, with entry by new patients at the earliest time that they had achieved both the age of 65 and 1 year of residence in Emilia‐Romagna for those immigrating into the region. The follow‐up time ended for individuals at the first of any of the following events: the end of the study period, emigration out of the region, hospitalization for more than 30 consecutive days or death.
Outcome
Hospital records were used to identify events for patients included in the study. The hospitalizations included as outcome events in the study were unplanned ordinary admissions that resulted in an overnight stay. The study evaluated all‐cause hospitalizations and no diagnoses were excluded.
Exposure
The Maio criteria are an Italian modified version of the Beers criteria described elsewhere 19, 23. The first version of the Maio criteria was developed in 2007 and it was subsequently revised to produce 2011 and 2014 versions (v2007, v2011 and v2014, respectively). The revisions were made to reflect changes in current research evidence, clinical practice, addition of new medications onto the market, and prescribers’ opinion and experience. The criteria consist of three distinct categories of PIMs: medications that ‘should always be avoided’, medications that are ‘rarely appropriate’ and medications that have ‘some indications’ for use in older adults but that are frequently misused. The category of medications that ‘should always be avoided’ is divided into those reimbursed by the Italian National Formulary, and therefore detectable in the pharmacy records, and those not reimbursed (the full list of the 2014 Maio criteria is available in Table S1). As the objective of the present study was to estimate the association between PIM exposure and hospitalization using the Emilia‐Romagna database, PIM exposure was defined as reimbursed PIMs that ‘should always be avoided’ as the net clinical effect of these medications is expected to be negative, regardless of indication. A comprehensive list of the reimbursed medications classified as ‘should always be avoided’ from each of the three versions of the Maio criteria can be found in Table 1.
Table 1.
PIM | Maio criteria v2007 | Maio criteria v2011 | Maio criteria v2014 |
---|---|---|---|
Amitriptyline | ✓ | ✓ | ✓ |
Chlorpropamide | ✓ | ✓ | ✓ |
Cimetidine | ✓ | ||
Clonidine (oral) | ✓ | ✓ | ✓ |
Digoxin >0.125 mg day –1 | ✓ | ||
Disopyramide | ✓ | ✓ | ✓ |
Ferrous sulphate >325 mg day –1 | ✓ | ||
Indomethacin | ✓ | ✓ | ✓ |
Ketorolac injectable (>2 days) | ✓ | ✓ | ✓ |
Methyldopa | ✓ | ✓ | ✓ |
Nifedipine (short‐acting) | ✓ | ✓ | ✓ |
Nitrofurantoin | ✓ | ||
NSAIDs oral (>15 days) | ✓ | ✓ | ✓ |
Oestrogens (oral) | ✓ | ✓ | ✓ |
Orphenadrine | ✓ | ✓ | ✓ |
Pentazocine | ✓ | ✓ | ✓ |
Testosterone | ✓ | ✓ | ✓ |
Spironolactone >25 mg day –1 | ✓ | ✓ | |
Escitalopram >10 mg day –1 | ✓ | ✓ | |
Citalopram >20 mg day –1 | ✓ | ✓ | |
Ticlopidine | ✓ | ||
Imipramine | ✓ | ||
Nortriptyline | ✓ | ||
Trimipramine | ✓ | ||
Clomipramine | ✓ |
NSAID, nonsteroidal anti‐inflammatory drug
Several important changes were implemented across the three versions of the Maio criteria. Digoxin was reclassified from medications that ‘should always be avoided’ in v2007 to medications that have ‘some indications’ in v2011 and v2014. This was done as a result of clinicians’ belief that the benefits of digoxin outweigh the risks in certain patients. Cimetidine was removed from the medications that ‘should always be avoided’ for v2011 and v2014 because of decreases in utilization. Ferrous sulphate was removed from the ‘always be avoided’ category because clinicians no longer felt that prescribing patterns exposed patients to substantial risks. Furthermore, nitrofurantoin was removed from the ‘always be avoided’ class because it is no longer covered by the Italian National Formulary and its utilization is no longer observable in the database. The dose‐related use of spironolactone, escitalopram and citalopram dosages in excess of specified thresholds were added to the ‘always be avoided‘ class, starting with the v2011 Maio criteria. Owing to an increase in use and an accumulation of evidence demonstrating an elevated risk of bleeding, ticlopidine was reclassified and added to the ‘always be avoided’ class in v2014. Finally, imipramine, nortriptyline, trimipramine and clomipramine were added to the ‘always be avoided’ class in the v2014 Maio criteria (Table 1).
Exposure to PIMs was defined for the present study using solely the class of medications that ‘should always be avoided’. These PIMs are described as either ineffective or as unnecessarily high risk for the elderly population, based on current evidence, and for which safer alternative medications might exist on the formulary 19. Each day of follow‐up was defined as exposed or unexposed for each patient. Exposure to PIMs was operationalized as the estimated number of days supplied for each medication of interest 19, plus 30 days. The 30‐day period following the last day of the supply period was included to reflect medication usage patterns in a real‐world setting with partial non‐adherence and medication accumulation. If a hospitalization event occurred during a period of exposure to at least one PIM, that event was associated with the PIM (i.e. a PIM‐related hospitalization).
Statistical methods
Sample characteristics were calculated using person‐time exposed to PIM vs. not exposed to PIM for each patient. This approach quantifies the amount of time that each individual was exposed and not exposed to a PIM in person‐years and represents the sample characteristics as the proportion of person‐time contributed by patients with a given characteristic and exposure status. This does not necessarily estimate the number of individuals that were ever exposed or not exposed to PIMs but does more accurately represent the relationships between study variables and the PIM exposure time used in rate estimates. Crude PIM‐related hospitalization rates were calculated for each individual PIM by dividing the total number of hospitalizations related to a given PIM by the cumulative exposure time (per 1000 person‐years of exposure) to that PIM in the cohort over the study period. Crude PIM‐related hospitalization rates were calculated in the same manner for each version of the Maio criteria.
The rate of hospitalization among those experiencing PIM exposure was compared with the rate of hospitalization among those not experiencing PIM exposure in the population using a repeated‐events Cox proportional hazards model with time‐dependent covariates, using the counting process style of input. A time‐dependent variable was used to represent PIM exposure, which can repeatedly ‘switch on’ and ‘switch off’ for an individual as their PIM exposure status changes over time. The model was adjusted for gender and time‐dependent covariates for age, the number of non‐PIM‐related hospitalizations in the previous four quarters and the number of chronic conditions in the previous four quarters. The number of chronic conditions was calculated based on the chronic condition drug groups (CCDGs), which uses the number of prescriptions to identify up to 31 specific chronic diseases 24. The number of CCDGs represents the number of distinct chronic conditions that an individual has had in the previous four quarters. This number and the number of non‐PIM‐related hospitalizations were updated in each quarter, for each individual, throughout the study period. Using the hazard ratio (HR) from the Cox proportional hazards model, the population attributable fraction of hospitalizations due to v2014 PIMs was estimated {attributable fraction = [prevalence of exposed person‐time*(HR–1)]/[prevalence of exposed person‐time*(HR–1)] + 1]} and this fraction was multiplied by the total number of hospitalizations to estimate the number of hospitalizations attributable to PIM.
Owing to the computational intensity of fitting this model to such a large dataset of repeated events and time‐dependent covariates, the model could not be fitted at one time. To address this, we randomly split the cohort up into 10 subsets with roughly equal numbers of individuals. The model was fitted to each of the 10 subsets to estimate 10 independent and identically distributed (iid) parameter estimates. Applying the central limit theorem and what we know about the distribution of the sum of iid normal random variates, we aggregated the 10 parameter estimates (i.e. the log HRs) for each model covariate by computing their means. The standard error for each of the aggregated parameter estimates was calculated by computing the mean of the 10 parameter variance estimates, dividing by 10 and taking the square root. These standard deviations were used to compute 95% confidence intervals (CIs) around the parameter estimates before antilogging the parameter and CI estimates to arrive at the HRs and their respective CIs in the whole cohort.
The Thomas Jefferson University Institutional Review Board determined that the study did not comprise human subjects research.
Results
Demographics
The study included a total of 1 480 137 individuals, with a median follow‐up period of 7.2 years, with interquartile range (IQR) [3.2, 10.9] (Table 2). A total of 802 815 (54.2%) individuals in the study were exposed to a PIM at least once. Out of the total of 10 004 367 person‐years of follow‐up, 1 094 682 person‐years were attributed with v2014 PIM exposure, accounting for 10.9% of the total time. Persons over the age of 75 contributed 49.6% of person‐time in the full study cohort; 31.4% of total follow‐up time in the cohort was contributed by individuals with four or more chronic conditions. The majority of exposed person‐time was contributed by patients older than 75 years (57.6%) and by those with four or more chronic conditions (54.7%). Nearly a fifth (18.6%) of exposed person‐time was contributed by those with a history of previous hospitalizations.
Table 2.
Exposure to v2007 PIM | Exposure to v2011 PIM | Exposure to v2014 PIM | Full cohort | |||||
---|---|---|---|---|---|---|---|---|
1000 PYs contributed | % | 1000 PYs contributed | % | 1000 PYs contributed | % | 1000 PYs contributed | % | |
Total | 789.8 | 100.0 | 767.2 | 100.0 | 1094.7 | 100.0 | 10004.4 | 100.0 |
Gender | ||||||||
Female | 531.7 | 67.3 | 548.1 | 71.4 | 705.6 | 64.5 | 5763.2 | 57.6 |
Male | 258.1 | 32.7 | 219.1 | 28.6 | 389.1 | 35.5 | 4241.2 | 42.4 |
Age (years) | ||||||||
65–74 | 315.9 | 40.0 | 344.1 | 44.9 | 464.3 | 42.4 | 5045.6 | 50.4 |
75–84 | 332.5 | 42.1 | 315.2 | 41.1 | 463.1 | 42.3 | 3630.7 | 36.3 |
85+ | 141.4 | 17.9 | 107.9 | 14.1 | 167.3 | 15.3 | 1328.1 | 13.3 |
Number of previous hospitalizations | ||||||||
0 | 660.5 | 83.6 | 643.4 | 83.9 | 890.7 | 81.4 | 8334.8 | 83.3 |
1+ | 129.30 | 16.4 | 123.7 | 16.1 | 204.0 | 18.6 | 1669.6 | 16.7 |
Number of chronic conditions a | ||||||||
0 | 5.7 | 0.7 | 5.7 | 0.7 | 6.2 | 0.6 | 883.9 | 8.8 |
1–3 | 375.4 | 47.5 | 338.9 | 44.2 | 490.2 | 44.8 | 5981.5 | 59.8 |
4+ | 408.7 | 51.8 | 422.6 | 55.1 | 598.3 | 54.7 | 3138.9 | 31.4 |
Chronic conditions are calculated based on the chronic condition drug groups 24, which uses the number of prescriptions to identify 31 specific chronic diseases
PIM, potentially inappropriate medication; PY, person‐years
Hospitalizations and PIM exposure
The number of hospitalizations which occurred while patients were exposed to one or more v2014 PIMs was 249 673 (15.6%) (Table 3). The crude rates of hospitalization during v2014 PIM‐exposed and unexposed person‐time were 228.1 and 152.1 hospitalizations per 1000 person‐years, respectively. Crude rates for individual PIMs ranged from a minimum of 57.4 (oestrogen) to a maximum of 1137.9 (ketorolac) hospitalizations per 1000 person‐years (Table 4). Besides oestrogen, the hospitalization rate during exposure to each of the other 24 individual PIMs was 155 hospitalizations per 1000 person‐years or greater. Patients were exposed for large quantities of person‐time to several newly added medications in v2014 of the Maio criteria, including ticlopidine, citalopram, escitalopram and spironolactone (340 500, 143 900, 88 900 and 17 800 person‐years, respectively) and hospitalization rates were elevated during periods of exposure to these medications (283.3, 269.8, 248.7 and 473.4 hospitalizations per 1000 person‐years, respectively). Substantial person‐time was classified as exposed to oral nonsteroidal anti‐inflammatory drugs (NSAIDs) (470 600 person‐years) and the hospitalization rate was 175.5 per 1000 person years during NSAID exposure.
Table 3.
Version of the Maio criteria | Exposure status | Number of people | Time in 1000 PYs | Number of hospitalizations | Hospitalization rate per 1000 PYs a |
---|---|---|---|---|---|
v2007 | Exposed | 769 191 | 789.8 | 196 480 | 248.8 |
v2007 | Unexposed | 1 477 094 | 9214.9 | 1 408 421 | 152.8 |
v2011 | Exposed | 741 167 | 767.2 | 161 682 | 210.7 |
v2011 | Unexposed | 1 476 890 | 9237.6 | 1 443 219 | 156.2 |
v2014 | Exposed | 802 815 | 1094.7 | 249 673 | 228.1 |
v2014 | Unexposed | 1 474 842 | 8909.7 | 1 355 228 | 152.1 |
Rate of hospitalization was calculated by dividing the number of hospitalizations by the exposure time
PIM, potentially inappropriate medication; PY, person‐years
Table 4.
Medication(s) | Number of people | Exposure time in 1000 PYs | Number of hospitalizations | Rate of hospitalization per 1000 PYs a |
---|---|---|---|---|
Ketorolac injectable (>2 days) | 11 992 | 1.6 | 1857 | 1137.9 |
Pentazocine | 91 | 0.0 | 18 | 525.1 |
Nitrofurantoin b | 1897 | 0.3 | 139 | 500.2 |
Ferrous sulphate >325 mg day –1 b | 141 468 | 108.9 | 52 313 | 480.2 |
Spironolactone > 25 mg day –1 c | 15 924 | 17.8 | 8440 | 473.4 |
Digoxin > 0.125 mg day –1 b | 84 071 | 163.0 | 54 221 | 332.6 |
Ticlopidine d | 151 758 | 340.5 | 96 463 | 283.3 |
Clonidine (oral) | 7349 | 11.8 | 3288 | 278.9 |
Citalopram > 20 mg day –1 c | 117 297 | 143.9 | 38 826 | 269.8 |
Chlorpropamide | 29 | 0.0 | 9 | 264.7 |
Orphenadrine | 5354 | 5.8 | 1459 | 252.5 |
Escitalopram > 10 mg day –1 c | 70 129 | 88.9 | 22 109 | 248.7 |
Amitriptyline | 45 358 | 32.1 | 7675 | 238.7 |
Indomethacin | 30 962 | 7.0 | 1585 | 225.2 |
Testosterone | 415 | 0.4 | 100 | 223.3 |
Nifedipine (short‐acting) | 1120 | 0.6 | 138 | 221.1 |
Cimetidine b | 885 | 0.9 | 177 | 190.0 |
Disopyramide | 714 | 1.2 | 227 | 183.0 |
NSAIDs (>15 days) | 633 433 | 470.6 | 82 580 | 175.5 |
Trimipramine d | 1788 | 1.8 | 311 | 170.5 |
Methyldopa | 1134 | 2.0 | 333 | 163.4 |
Nortriptyline d | 2648 | 4.3 | 686 | 158.5 |
Imipramine d | 2046 | 2.2 | 341 | 155.7 |
Clomipramine d | 9075 | 14.8 | 2295 | 155.1 |
Oestrogen (oral) | 5939 | 7.5 | 432 | 57.4 |
NSAID, nonsteroidal anti‐inflammatory drug; PIM, potentially inappropriate medication; PY, person‐years
Rate of hospitalization was calculated by dividing the number of hospitalizations by the exposure time
Always avoided PIMs in the 2007 version but not in the 2011 and 2014 versions of the Maio criteria
Always avoided PIMs in the 2011 and 2014 versions but not in the 2007 version of the Maio criteria
Always avoided PIMs in the 2014 version but not in the 2007 and 2011 versions of the Maio criteria
Multivariable analysis
In the multivariable repeated events Cox proportional hazard model, the hazard of hospitalization was 16% greater (HR = 1.16, 95% CI 1.14, 1.18) during exposure to the v2014 Maio criteria PIMs (Table 5). The HR estimates across different versions of the Maio criteria were directionally aligned and of similar magnitude (v2007: HR = 1.24, 95% CI 1.24, 1.25; v2011: HR = 1.12, 95% CI 1.12, 1.13).
Table 5.
Version of the Maio criteria | HR for PIM exposure | 95% CI |
---|---|---|
v2007 | 1.24 | (1.24, 1.25) |
v2011 | 1.12 | (1.12, 1.13) |
v2014 | 1.16 | (1.14, 1.18) |
CI, confidence interval; HR, hazard ratio; PIM, potentially inappropriate medication
Discussion
The hazard of hospitalization was increased during exposure to PIMs in this large population‐based cohort of 1 480 137 older adults with more than 10 million years of observation and nearly 1.1 million years of PIM exposure. The 16% increase in the hazard of hospitalization during exposure to medications classified as ‘should always be avoided’ by the v2014 Maio criteria was largely consistent with 24% and 12% increased hazards during exposure to medications classified as ‘should always be avoided’ in the v2007 and v2011 criteria, respectively. Crude rates of hospitalization were consistently high during exposure to individual PIMs.
Effect estimates for PIM exposure were largely consistent across the three versions of the Maio criteria. Digoxin rescheduling in the 2011 version of the Maio criteria was reflected in the notable decrease from 24% to 12% greater risk during PIM exposure as defined by the 2007 and 2011 versions, respectively. This decrease was observed despite the additions of three new medications with high hospitalization rates in 2011 (citalopram, escitalopram and spironolactone), as digoxin is a highly utilized medication in the elderly population with a high potential for harm. The risk with PIM exposure increased from 12% for v2011 to 16% after the addition of ticlopidine in v2014. It is unsurprising that this drug was implicated in a large number of hospitalizations. Ticlopidine is a highly utilized antiplatelet agent in Emilia‐Romagna, and haematological agents have been reported as the source of 42% of adverse drug event (ADE)‐related hospitalizations in the United States 16. Our findings for ticlopidine are comparable with those of an earlier study which reported an adjusted odds ratio for unplanned hospitalization of 1.50 for ticlopidine exposure 25. The NSAID class accounted for a sizable amount of exposed person‐time, and the crude rate of hospitalization during NSAID exposure was numerically greater than the crude rate during PIM unexposed time, in line with prior research documenting high utilization and increased risk with NSAIDs 3, 25.
Previous research
The international body of literature supporting the association between PIM exposure and hospitalization continues to expand 3, 12, 13, 14, 15, 25. A large segment of this research has applied intent‐to‐treat designs and estimated HRs for hospitalization over a duration of follow‐up for PIM‐exposed patients compared with patients either unexposed or exposed to a PIM alternative at each individual's index time point. With such approaches, an initial exposure to a PIM is hypothesized to precipitate a near‐term adverse event and hospitalization, or instead to initiate a cascade of events leading to subsequent adverse health outcomes observed during the course of follow‐up. It is important to consider the mechanism and latency period underlying PIM risk in the context of high variability in medication usage patterns over time in real‐world settings. In the majority of patients, PIM use will not produce an immediate adverse outcome, and the patient's altered risk due to PIM exposure is anticipated to return to baseline once the PIM is removed. Thus, our study's ‘as‐treated’ design is instructive and complementary to previous ‘intent‐to‐treat’ studies of PIM exposure and hospitalization.
Endres et al. analysed health insurance claims for nearly 400 000 German patients exposed to a PIM or PIM alternative, as defined by the PRISCUS list 13. The adjusted HR in the primary 180‐day all‐cause hospitalization analysis indicated a 38% increased hazard of hospitalization for PIM‐exposed patients. More than half (58%) of patients were exposed to PIM for one quarter or less 13, highlighting the temporal distancing between exposure and outcome with extended intent‐to‐treat follow‐up. A similar claims‐based study of nearly 50 000 Swiss patients enrolled in managed care plans estimated the risk of hospitalization in the year following an incident PIM dispensing as defined by the PRISCUS list 14. Compared with unexposed patients, the risk of hospitalization increased with the quantity of PIM exposures during the observation year from a HR of 1.13 for a single PIM to 1.63 for more than three PIMs 14. A dose–response relationship was also reported in a study by Price et al., with increasing odds ratios for unplanned hospitalization observed for each additional PIM exposure and with increasing volumes of defined daily doses 25.
In another study of approximately 150 000 commercially insured managed care enrollees in the United States, PIM exposure, as defined by the 2012 Beers criteria, was associated with an increased risk of ADEs, emergency department visits and hospitalization (HR: 2.17, 2.00, 2.03, respectively) in the month following the PIM exposure 12. Similar estimates were obtained when defining PIM exposure using either the Beers 2003 or STOPP criteria. The primary analytical approach applied a time‐dependent exposure definition, although the classification of exposure occurred at the month level. The overall hospitalization rate during the approximately 2.5 years of follow‐up (67 per 1000 person‐years) was lower than during the unexposed time in the Emilia‐Romagna cohort (152 per 1000 person‐years), potentially suggestive of population level health‐status and health system differences 12.
Based upon hospital surveillance data, nearly 100 000 hospitalizations have been estimated to be attributable to ADEs in the United States each year, representing roughly 0.3% of the 35 million short‐stay hospitalizations occurring annually 16. Remarkably, inappropriate medications, as classified by the Beers criteria, were implicated in only 3.2% (6.6% if digoxin was included) of the ADE hospitalizations, roughly 0.01% of the 35 million short‐stay hospitalizations 16. Although these values have been acknowledged to be underestimates due to limited exposure data and reliance on emergency physician documentation within clinical notes for outcome ascertainment, we also cannot definitively ascertain what proportion of the 15.6% of hospitalizations occurring during PIM exposure in our study was directly attributable to PIM. Under the theoretically ideal scenario where zero PIM exposures occur, there would have been 27 444 fewer hospitalizations. This represents an attributable fraction of 11% of 249 673 PIM‐exposed hospitalizations and 1.7% of all ~1.6 million hospitalizations in Emilia‐Romagna during the study period. This estimate is comparable with an attributable fraction of 15% of PIM‐exposed hospitalizations estimated in a longitudinal study of Western Australian patients 25.
Strengths and limitations
The present study examined PIM exposure for a population‐level cohort of more than 1 million patients for more than a decade of observation. The universal healthcare system minimized loss to follow‐up and facilitated near‐complete visibility to exposure and outcome events, including for NSAIDs, a medication class typically unobserved in claims owing to their availability over the counter. Time‐dependent exposure classification throughout the study ensured a strong temporal association between PIM use and outcome events, while also permitting the calculation of incidence rates. Time‐dependent covariate definitions for comorbid disease burden and prior hospitalizations mitigated potential sources of time‐varying confounding. The evaluation of PIMs, as defined by all three generations of the Maio criteria, provided insight into the impact of changes to the criteria.
Limitations were consistent with those of a claims‐based retrospective cohort study. The use of pharmacy dispensing data to determine medication exposure relies on the assumption that patients who fill their prescriptions also take them as prescribed. An important limitation was the inability to determine whether any individual hospitalization was truly caused by PIM exposure. However, the close temporal association between exposure and hospitalization, the adjustment for prominent time‐dependent confounders, and alignment with prior research supports the validity of the increased hazard of hospitalization reported during PIM exposure. Intuitively, the relationship between PIMs and hospitalization is driven by the way that PIMs are defined. Certain medications, most prominently long‐acting benzodiazepines, are classified on the full Maio criteria as ‘should always be avoided’ but dispensings are not detectable in the Emilia‐Romagna pharmacy database because these medications are not reimbursed by the Italian National Formulary. As long‐acting benzodiazepines have been estimated to be associated with an increased hospitalization risk of 16–27%, depending upon the medication 25, our findings may represent an underestimate of the risk associated with PIM exposure. Finally, additive effects of multiple concurrent PIM exposures were not examined; this is an area requiring further investigation.
Conclusion
In a large population‐based cohort of older adults, we found a sizeable increase in the hospitalization rate during exposure to PIMs, as defined by each of the three versions of the Maio criteria. With accumulating evidence suggesting an increased hospitalization risk with PIM exposure, it has become critically important to understand the risks associated with particular drug classes and individual medications. A thorough characterization of the differences in risk between patient subgroups and across medications is needed to guide clinical practice meaningfully. As was the case with digoxin, certain medications considered to be potentially inappropriate will have a net benefit–risk profile supporting their use in certain older adults in the context of patient‐centred care and precision medicine. Despite such exceptions, in aggregate, the prevalence of PIM exposure remains remarkably high. Thus, it remains imperative that researchers produce instructive evidence to influence prescribers, while informing ongoing expansion and revision of various PIM criteria. Interventions designed to disseminate evidence, educate clinicians and reduce PIM prescribing are needed to achieve the overarching goal of minimizing the burden of PIM‐related ADEs for patients and the healthcare system.
Competing Interests
There are no competing interests to declare.
Data for this study was retrieved from the regional database of the Emilia‐Romagna Region, provided through a collaborative agreement between the Regional Health Care and Social Agency, Emilia‐Romagna, Italy; the Health Care Authority, Emilia‐Romagna, Italy; and Thomas Jefferson University, Philadelphia, PA, USA. The collaborative agreement aimed at supporting the availability of data for the implementation of a research project in the Parma Local Health Authority to promote drug prescription appropriateness.
Contributors
V.M, acquired the data. V.M., S.V., M.A., S.K., S.D.C., M.L. and S.H. designed the study. S.H. and S.K. analysed the data. S.V. and M.A. drafted the manuscript. V.M., S.D.C., M.L., S.K. and S.H. provided critical revisions to the manuscript.
Supporting information
Varga, S. , Alcusky, M. , Keith, S. W. , Hegarty, S. E. , Del Canale, S. , Lombardi, M. , and Maio, V. (2017) Hospitalization rates during potentially inappropriate medication use in a large population‐based cohort of older adults. Br J Clin Pharmacol, 83: 2572–2580. doi: 10.1111/bcp.13365.
References
- 1. Amos TB, Keith SW, Del Canale S, Orsi P, Maggio M, Baccarini S, et al. Inappropriate prescribing in a large community‐dwelling older population: a focus on prevalence and how it relates to patient and physician characteristics. J Clin Pharm Ther 2015; 40: 7–13. [DOI] [PubMed] [Google Scholar]
- 2. Davidoff AJ, Miller GE, Sarpong EM, Yang E, Brandt N, Fick DM. Prevalence of potentially inappropriate medication use in older adults using the 2012 beers criteria. J Am Geriatr Soc 2015; 63: 486–500. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Narayan SW, Nishtala PS. Associations of potentially inappropriate medicine use with fall‐related hospitalizations and primary care visits in older New Zealanders: a population‐level study using the updated 2012 beers criteria. Drugs Real World Outcomes 2015; 2: 137–141. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Morin L, Fastborn J, Laroche ML, Johnell K. Potentially inappropriate drug use in older people: a nationwide comparison of different explicit criteria for population‐based estimates. Br J Clin Pharmacol 2015; 80: 315–324. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Erdem E. Prevalence of chronic conditions among Medicare part a beneficiaries in 2008 and 2010: are Medicare beneficiaries getting sicker? Prev Chronic Dis 2014; 11: 130118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Charlesworth CJ, Smit E, Lee DS, Alramadhan F, Odden MC. Polypharmacy among adults aged 65 years and older in the United States: 1988–2010. J Gerontol A Biol Sci Med Sci 2015; 70: 989–995. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Lamy PP. Physiological changes due to age. Drugs Aging 1991; 5: 385–404. [DOI] [PubMed] [Google Scholar]
- 8. Beers MH. Explicit criteria for determining potentially inappropriate medication use by the elderly. An update. Arch Intern Med 1997; 157: 1531–1536. [PubMed] [Google Scholar]
- 9. Corsonello A, Onder G, Abbatecola AM, Guffanti EE, Gareri P, Lattanzio F. Explicit criteria for potentially inappropriate medications to reduce the risk of adverse drug reactions in elderly people. Drug Saf 2012; 35: 21–28. [DOI] [PubMed] [Google Scholar]
- 10. Keith SW, Maio V, Dudash K, Templin M, Del Canale S. A physician‐focused intervention to reduce inappropriate medications prescribing to older people. Drugs Aging 2013; 30: 119–127. [DOI] [PubMed] [Google Scholar]
- 11. Lopatto J, Keith SW, Del Canale S, Templin M, Maio V. Evaluating sustained quality improvements: long‐term effectiveness of a physician‐focused intervention to reduce potentially inappropriate medication prescribing in an older population. J Clin Pharm Ther 2014; 39: 266–271. [DOI] [PubMed] [Google Scholar]
- 12. Brown JD, Hutchison LC, Li C, Painter JT, Martin BC. Predictive validity of the beers and screening tool of older Persons' potentially inappropriate prescriptions (STOPP) criteria to detect adverse drug events, hospitalizations, and emergency department visits in the United States. J Am Geriatr Soc 2016; 64: 22–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Endres HG, Kaufmann‐Kolle P, Steeb V, Bauer E, Bottner C, Thurmann P. Association between potentially inappropriate medication (PIM) use and risk of hospitalization in older adults: an observational study based on routine data comparing PIM use with use of PIM alternatives. PLoS One 2016; 11: e0146811. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Reich O, Rosemann T, Rapold R, Blozik E, Senn O. Potentially inappropriate medication use in older patients in Swiss managed care plans: prevalence, determinants and association with hospitalization. PLoS One 2014; 9: e105425. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Wauters M, Elseviers M, Vaes B, Degryse J, Dalleur O, Vander Stichele R, et al. Too many, too few, or too unsafe? Impact of inappropriate prescribing on mortality, and hospitalization in a cohort of community‐dwelling oldest old. Br J Clin Pharmacol 2016; 82: 1382–1392. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Budnitz DS, Lovegrove MC, Shehab N, Richards CL. Emergency hospitalizations for adverse drug events in older Americans. N Engl J Med 2011; 365: 2002–2012. [DOI] [PubMed] [Google Scholar]
- 17. Pasina L, Djade CD, Tettamati M, Franchi C, Salerno F, Corrao S, et al. REPOSI investigators. Prevalence of potentially inappropriate medications and risk of adverse clinical outcome in a cohort of hospitalized elderly patients: results from the REPOSE study. J Clin Pharm Ther 2014; 39: 511–515. [DOI] [PubMed] [Google Scholar]
- 18. Shehab N, Lovegrove MC, Geller AI, Rose KO, Weidle NJ, Budnitz DS. US emergency department visits for outpatient adverse drug events, 2013–2014. JAMA 2016; 316: 2115–2125. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Maio V, Del Canale S, Abouzaid S. GAP investigators. Using explicit criteria to evaluate the quality of prescribing in elderly Italian outpatients: a cohort study. J Clin Pharm Ther 2010; 35: 219–229. [DOI] [PubMed] [Google Scholar]
- 20. Lo Scalzo A, Donatini A, Orzella L, Cicchetti A, Profili S, Maresso A, et al. European Observatory on Health Care Systems and Policies . Italy: health system review. Health Syst Transit 2009; 11: 1–243. [Google Scholar]
- 21. Squires D. The Commonwealth Fund . International profiles of health care systems: the Italian health care system, 2009. June 2010; pub. no. 1417: 32–5.
- 22. World Health Organization . Guidelines for ATC classification and DDD assignment 2013. 2013. Available at http://www.whocc.no/filearchive/publications/1_2013guidelines.pdf (last accessed 29 April 2016).
- 23. Kaufmann CP, Tremp R, Hersberger KE, Lampert ML. Inappropriate prescribing: a systematic overview of published assessment tools. Eur J Clin Pharmacol 2014; 70: 1–11. [DOI] [PubMed] [Google Scholar]
- 24. Maio V, Yuen E, Rabinowitz C, Louis D, Jimbo M, Donatini A, et al. Using pharmacy data to identify those with chronic conditions in Emilia Romagna, Italy. J Health Serv Res Policy 2005; 10: 232–238. [DOI] [PubMed] [Google Scholar]
- 25. Price SD, Holman CD, Sanfilippo FM, Emery JD. Association between potentially inappropriate medications from the beers criteria and the risk of unplanned hospitalization in elderly patients. Ann Pharmacother 2014; 48: 6–16. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.