Predictors of a Gabapentinoid–Loop‐Diuretic Prescribing Cascade in U.S. Nursing Home Residents

ABSTRACT

Background

Gabapentinoid‐related peripheral edema may prompt loop diuretic prescribing. Nursing home (NH) residents may be especially prone to this prescribing cascade. We estimated the incidence and identified predictors of the gabapentinoid‐loop diuretic prescribing cascade in NHs.

Methods

We conducted a retrospective cohort study using 2016–2022 Medicare claims linked with Minimum Data Set assessments. We identified residents aged ≥ 66 years who initiated gabapentinoids in NHs and who had no evidence of loop diuretic use, heart failure, or renal insufficiency during the prior 6 months. The outcome was loop diuretic initiation within 90 days of gabapentinoid initiation. Using multivariable Poisson regression models, we estimated adjusted risk ratios (aRR) with 95% robust confidence intervals to identify predictors. We used pooled logistic regression models to examine the relationship between time‐varying gabapentinoid dose and loop diuretic initiation risk.

Results

Among 23,544 residents, 994 (4.2%) experienced a prescribing cascade at a median of 36 days (IQR 15–61) after gabapentinoid initiation. Risk was higher with age 86–90 years (aRR = 1.60) or ≥ 91 years (aRR = 1.38); a diagnosis of chronic pain or fibromyalgia (aRR = 1.16), or diabetes (aRR = 1.23); and receipt of potassium‐sparing diuretics (aRR = 1.53), thiazide diuretics (aRR = 1.27), or 15 or more unique medications (aRR = 1.18). Higher (versus lower) weekly gabapentin dose during follow‐up was associated with a 1.45 times higher prescribing cascade risk over 13 weeks. Those with Alzheimer's Disease and Related Dementias (aRR = 0.79), or moderate (aRR = 0.72) to severe cognitive impairment (aRR = 0.59) had a lower risk versus those with intact cognition.

Conclusions

Approximately 1 in 20 NH residents who initiate gabapentinoids receives a loop diuretic within 3 months. Potentially modifiable predictors included existing polypharmacy and titrating gabapentinoid doses. NH clinicians should monitor for edema soon after gabapentinoid initiation and consider dose reductions or discontinuation before adding a loop diuretic.

Summary

• Key Points

  • ○
    Among 23,544 older adult long‐stay nursing home residents who initiated a gabapentinoid in the nursing home (2016–2022), 4.2% experienced a loop diuretic prescribing cascade within 90 days.
  • ○
    Predictors of the prescribing cascade that are potentially intervenable included a higher medication burden (≥ 15 medications), current receipt of non‐loop diuretics, and titration of gabapentinoid dose after initiation
  • ○
    Other predictors associated with increased risk included older age, greater comorbidity, and a diagnosis of neuropathic pain, fibromyalgia, or diabetes, while more severe cognitive impairment or a diagnosis of Alzheimer's Disease and Related Dementias was associated with lower risk.
• Why does this paper matter?

  • ○
    We found evidence of a higher incidence of a gabapentinoid‐loop diuretic prescribing cascade among long‐stay nursing home residents, as compared to prior estimates from community‐dwelling older adult cohorts.
  • ○
    Significant predictors in our analysis reflected a higher propensity to experience edema.
  • ○
    NH clinicians initiating a gabapentinoid should consider intervening on modifiable predictors, monitoring for edema especially in those with non‐modifiable predictors, and considering dose reductions or discontinuation before adding a loop diuretic.

Among 23,544 older adult long‐stay nursing home residents who initiated a gabapentinoid in the nursing home (2016–2022), 4.2% experienced a loop diuretic prescribing cascade within 90 days. Predictors of the prescribing cascade included older age, higher medication burden (≥ 15 medications), current receipt of non‐loop diuretics, and titration of gabapentinoid dose after initiation.

1. Introduction

A prescribing cascade occurs when a medication's adverse effect results in the prescription of an additional and potentially unnecessary medication to treat the symptom [1, 2]. This process contributes to polypharmacy and increases the risk of adverse drug reactions, functional decline, and healthcare utilization [3, 4]. One example is the gabapentinoid‐loop diuretic prescribing cascade. Peripheral edema is a known side effect of gabapentinoids (gabapentin or pregabalin), with an estimated incidence of 7.6% to 16.0% depending on the population, indication, and drug [5, 6, 7]. Clinicians may prescribe a loop diuretic (e.g., furosemide, bumetanide) to manage the symptom [8]. This prescribing cascade can lead to complications, including electrolyte imbalances, hypotension, and increased fall risk [9, 10].

Gabapentinoids are antiepileptic medications approved to manage seizures and treat neuropathic pain [11, 12]. In nursing homes (NH) 13.3% of long‐stay residents received gabapentin and 1.8% received pregabalin in 2016 [13, 14]. According to a recent survey of 187 NH clinicians, many viewed gabapentinoids as low‐risk for various off‐label uses [15], especially since these medications are not part of federal quality measures and receive less scrutiny than opioids or antipsychotics [16]. The growing list of off‐label uses, such as non‐neuropathic pain, agitation, insomnia, alcohol withdrawal, and anxiety, likely contributes to increased prescribing and therefore to the risk of loop‐diuretic prescribing cascades. Although previous research has documented the gabapentinoid‐loop diuretic prescribing cascade in older adults [8], no study has specifically examined it in long‐term NH residents. This is a critical gap in knowledge since NH residents are uniquely vulnerable to potentially inappropriate medication use [17, 18] and prescribing cascades [19].

Multimorbidity and polypharmacy are common in NH residents [20, 21], and increase the likelihood of adverse drug effects and complex medication regimens [22, 23]. Cognitive impairment from Alzheimer's disease and related dementias (ADRD) is common [24] and may limit residents' ability to report medication side effects accurately. Aspects of NH care systems may also increase the likelihood of prescribing cascades. Staff turnover [15, 25] and missing documentation during frequent care transitions [15, 25] can make it difficult to recognize adverse drug effects. Given these unique factors, it is important to understand if and how gabapentinoid‐loop diuretic prescribing cascades occur in NHs. This descriptive study aimed to estimate the incidence of and identify predictors for the gabapentinoid‐loop diuretic prescribing cascade within the NH setting in long‐term care residents initiating a gabapentinoid. Based on prior research in other populations, we hypothesized that men [26], residents prescribed pregabalin or higher gabapentinoid doses [26, 27], those with hyperpolypharmacy [8, 28], and those with dementia [27] would be at higher risk of experiencing a gabapentinoid‐loop diuretic prescribing cascade, whereas residents aged ≥ 85 years may be at lower risk [8]. Other novel covariates such as prescriber specialization and characteristics of the facility were also explored.

2. Methods

2.1. Data Sources

We leveraged Medicare enrollment and claims data, Minimum Dataset (MDS) clinical assessments, the Medicare Data on Provider Practice and Specialty (MD‐PPAS) file, and the Certification and Survey Provider Enhanced Reports (CASPER) dataset. The Medicare Beneficiary Summary File provided information on enrollment, beneficiary demographics, date of death, and chronic conditions (i.e., the Chronic Conditions Warehouse file) [29]. The Medicare Provider Analysis and Review (MedPAR) file contained inpatient and skilled nursing facility claims (similar to Part A claims) and diagnoses; Medicare Part B carrier claims contained information on ambulatory provider visits (e.g., emergency department) and diagnoses, and Part D provided prescription drug claims. We also leveraged the Residential History File to determine locations and transitions of care [30]. The MDS contains mandated clinical assessments conducted at least quarterly for residents in Medicare‐certified NHs. The MD‐PPAS file provided information on prescriber characteristics and the number of services delivered in different settings. The Brown University Institutional Review Board approved the study and waived informed consent due to the use of deidentified secondary data.

2.2. Study Design and Population

We conducted a retrospective cohort study. The target population of inference was older long‐term care NH residents who initiated gabapentinoids and were at risk for a loop diuretic prescribing cascade, defined as having no prior loop diuretic use or clinical indication for loop diuretics in the 6 months before gabapentinoid initiation. Because Medicare Part D claims do not capture medication use during periods of consolidated Part A billing for post‐acute care in NHs, we used an algorithm leveraging MDS assessments and Medicare claims to identify person‐time during which Medicare Part D claims would be generated (i.e., after post‐acute care), which we refer to as “drug‐observable NH episodes” [31]. Methodological details on the algorithm are available in Supporting Information 1.

We restricted these episodes to periods during which the residents were continuously enrolled in fee‐for‐service Medicare Parts A, B, and D. We identified residents who had their first gabapentinoid dispensing during a drug‐observable NH episode between 03/30/2016 and 09/30/2022. The date of the first gabapentinoid dispensing served as the cohort entry (index) date. We then excluded residents < 66 years or who were not continuously enrolled in Medicare Parts A, B, and D for the 12 months prior to the start of the NH episode. Because 30 days is the most common days' supply for gabapentinoid prescriptions and drug data are not observable during post‐acute care, we required that observable drug data span at least 45 days before the first gabapentinoid dispensing. This ensured a sufficient washout window to identify incident gabapentinoid use. Additional eligibility criteria were applied relative to the cohort entry date. These included not having any gabapentinoid or loop diuretic dispensings in any setting during the 180 days prior, at least one comprehensive or quarterly MDS assessment in the 120 days prior, and no documentation of a diagnosis of heart failure or renal insufficiency in claims or MDS assessments during that same period (diagnosis codes and MDS items used for exclusions provided in Table S1). We also excluded residents whose initial gabapentinoid days' supply was more than 30 days or who received the drug as a solution due to uncertainty in calculating daily dosages. Finally, we restricted the cohort to residents with complete covariate information for the Cognitive Function Scale (CFS) [32] and pain, extending the MDS assessment lookback window to 180 days before the cohort entry date. Figure S1 shows the study design and eligibility criteria.

2.3. Outcome and Follow‐Up

The primary outcome was the occurrence of the gabapentinoid‐loop diuretic prescribing cascade, defined as an incident dispensing of a loop diuretic (bumetanide, furosemide, torsemide, or ethacrynic acid) after initiation of a gabapentinoid. Thus, among all long‐term care residents who initiated a gabapentinoid, our comparison of interest was those who did versus did not subsequently start a loop diuretic during follow‐up. Follow‐up continued for 90 days or until the earliest of the following: loop diuretic dispensing; switching or discontinuing the gabapentinoid for > 7 days based on days supply; discharge from NH (including hospitalizations); disenrollment from Medicare Parts A, B, or D; or death. For all censoring criteria except disenrollment, censoring was considered to be a true absence of the prescribing cascade. If a resident disenrolled, the outcome was considered to be missing because it was not possible to determine whether the prescribing cascade occurred.

2.4. Covariates

We prespecified covariates of interest as potential predictors of the prescribing cascade based on prior literature [8, 26, 27, 28] and clinical knowledge. Covariates included sociodemographic characteristics (age, sex, and race/ethnicity [33]), clinical characteristics (the Combined Comorbidity Score ranging from −2 to 26 [34], the CFS ranging from 1 for no cognitive impairment to 4 for severe impairment [32], pain severity/frequency quality indicator [35], and selected chronic conditions obtained from the Chronic Condition Warehouse such as dementia and depression). We also measured gabapentinoid characteristics, including the drug prescribed (gabapentin or pregabalin) and lower vs. higher baseline daily dose. Dose was calculated as (dose*quantity)/days supply, with higher dose defined as ≥ 600 mg gabapentin or ≥ 150 mg pregabalin per day based on the distribution in the study population. We classified the number of unique drugs as 0–14 or ≥ 15, basing the cutoff on the univariate association of the number of drugs with the outcome. Additional covariates included selected drug classes; healthcare utilization; prescriber characteristics from the MD‐PPAS [36, 37] and facility characteristics. Detailed definitions for covariates are provided in Table S1.

2.5. Statistical Analyses

Poisson regression models were used to estimate unadjusted and adjusted risk ratios (aRRs) for the relationships between covariates and the prescribing cascade. We computed 95% confidence intervals (95% CI) for each risk ratio using robust standard errors. Throughout, assessments of statistical significance are made at the 5% significance level. To examine the correlations between cascade occurrence and dose changes over time, we then restructured the data to a person‐week dataset (i.e., a weekly discrete‐time “long” dataset) and fitted a pooled logistic regression model to estimate the cumulative incidence of a prescribing cascade while adjusting for baseline covariates and time‐varying dose of gabapentinoids. We specified the model to include an interaction between week and gabapentinoid dose to allow the association between dose and the cascade to vary by week and plotted cumulative incidence of the outcome by dose. These curves were marginalized over the empirical distributions of baseline covariates in the overall study population. All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, US) and Stata (16.1, StataCorp LLC, College Station, TX).

2.6. Stability Analyses

First, to explore the impact of missing outcomes due to Part D disenrollment during follow‐up, we conducted a stability analysis that classified all residents with missing outcomes as having experienced a prescribing cascade. This analysis reflects a worst‐case imputation scenario to evaluate the stability of our findings to potential informative censoring. Next, while our primary analysis used prespecified predictors, we also applied Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression to explore which variables were most strongly associated with the prescribing cascade using a data‐driven approach [38]. We included all prespecified predictors and used the Bayesian Information Criterion (BIC) to guide variable selection [39, 40]. We fit 2 LASSO iterations. The first stopped selecting variables after reaching the local minimum of BIC minimization to generate a parsimonious model. The second continued selection until the lowest BIC was reached to identify the overall best‐fitting model. We then entered the selected variables from each LASSO model into a subsequent Poisson model to estimate risk ratios and 95% CIs.

3. Results

3.1. Study Population

Among 368,065 drug‐observable NH episodes for 249,221 unique residents, we identified 23,544 eligible episodes with NH residents initiating a gabapentinoid (71.3% female, 81.8% non‐Hispanic White, median age 82 [Q1 = 75, Q3 = 89] years). A detailed exclusion flow diagram is available in Figure S2. Select characteristics are presented in Table 1, with all measured characteristics in Table S2. The median length of NH stay was 244 days. A total of 81.4% had a diagnosis of Alzheimer's disease or related dementias (ADRD), 58.1% had chronic pain or fibromyalgia, and 12.9% had epilepsy or convulsions. Gabapentin was the most commonly prescribed gabapentinoid (94.9%), and most residents (91.3%) received a lower dose at baseline. There was extensive polypharmacy in the prior 12 months, with 39.0% of residents dispensed 15 or more drugs. Residents had a mean of 3.7 unique prescribers of their medications, and most gabapentinoid prescribers were either family physicians or internists.

TABLE 1.

Characteristics of the study population of nursing home residents who initiated gabapentinoids in 2016–2022.

Characteristic a	Total N
	N (%) (unless otherwise specified)
Demographics
Age, years, median (Q1, Q3)	82 (75, 89)
Sex, n (%)
Female	16,789 (71.31)
Male	6755 (28.69)
Gabapentinoid characteristics
Gabapentinoid initiated
Gabapentin	22,339 (94.89)
Pregabalin	1205 (5.12)
Dose at Baseline, n (%)
Lower < 600 mg gabapentin or < 150 mg pregabalin per day	21,495 (91.30)
Higher ≥ 600 mg gabapentin or ≥ 150 mg pregabalin per day	2049 (8.70)
Combined comorbidity score, mean (SD) b , c	5.92 (3.10)
Cognitive Function Scale d
Cognitively intact	9024 (38.33)
Mild cognitive impairment	5984 (25.42)
Moderate cognitive impairment	7046 (29.93)
Severe cognitive impairment	1490 (6.33)
Co‐existing conditions, n (%) c
Anxiety and fear‐related disorders	16,170 (68.68)
Bipolar and Depressive related disorders	20,026 (85.06)
Chronic pain and fibromyalgia	13,677 (58.09)
ADRD	19,168 (81.41)
Diabetes	13,093 (55.61)
Epilepsy or convulsions	3034 (12.89)
No. of medications, mean (SD) c	13.49 (6.45)
Polypharmacy
No (0–14 drugs)	14,357 (60.98)
Yes (15+ drugs)	9187 (39.02)
Healthcare utilization c
No. of emergency department visits, mean (SD)	1.48 (2.17)
No. of hospitalizations, mean (SD)	0.71 (1.11)
Length of NH stay at index date‐ days, Median (Q1, Q3)	244 (121, 491)
Prescribers
No. of unique clinicians (prescribers) who prescribed a medication, mean (SD)	3.69 (2.47)
Pain severity/frequency, n (%) d
None	13,103 (55.65)
Mild/infrequent	7019 (29.81)
Severe/frequent	3422 (14.53)

Abbreviations: ADRD, Alzheimer's disease and related dementias; Q1, first quartile; Q3, third quartile; SD, standard deviation; SNRI, serotonin–norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor.

^a All the characteristics were summarized at the level of the first gabapentinoid initiation.

^b The Gagne combined comorbidity score was calculated. Its values range from a minimum −2 to a maximum of 26.

^c Measured over the 365 days before gabapentinoid initiation.

^d Measured over the 120 days before gabapentinoid initiation, up to 180 days before, if missing within 120 days.

3.2. Incidence and Predictors of the Gabapentinoid‐Loop Diuretic Prescribing Cascade

By the end of the 90‐day follow‐up period, 4.2% (n = 994) of the study population experienced a gabapentinoid‐loop diuretic prescribing cascade. The median number of days between gabapentinoid initiation and first dispensing of loop diuretics was 28 (Q1 = 13, Q3 = 54) days. Most residents who did not experience the outcome were censored at the end of the 90‐day follow‐up period (54.7%), followed by gabapentin discontinuation (32.9%), and death (6.0%).

Select aRRs with 95% CIs for previously hypothesized variables are shown in Figure 1. All unadjusted RRs and aRRs with 95% CIs are presented in Table S3. Nonmodifiable predictors of the gabapentinoid‐loop diuretic cascade included age, chronic pain/fibromyalgia (aRR = 1.16 [95% CI = 1.01–1.33]), and diabetes (aRR = 1.23 [1.08, 1.40]). Residents aged 86–90 years and those aged 91 or older had 1.60 times [1.25, 2.04] and 1.38 times [1.06, 1.79] the adjusted risk, respectively, compared with residents aged 66–70 years. In contrast, those with a diagnosis of ADRD had a lower risk of loop‐diuretic dispensing (aRR = 0.79 [0.68, 0.93]). Moderate and severe cognitive impairment were also associated with lower risk compared to intact cognition (aRR = 0.72 [0.61, 0.85]) and (aRR = 0.59 [0.41, 0.84]), respectively. Later years were also associated with lower risk (2020 vs. 2018 aRR = 0.71 [0.56, 0.91]; and 2022 vs. 2018 aRR = 0.72 [0.55, 0.95]). None of the prescriber or facility characteristics were significantly associated with the risk of having the prescribing cascade.

FIGURE 1.

Selected adjusted risk ratios for predictors of the gabapentinoid‐loop diuretic prescribing cascade among nursing home residents. **denotes statistically significant adjusted risk ratio. All models adjusted for age, sex, race and ethnicity, calendar year, type of gabapentinoid and dose initiated, co‐medication use, comorbidities, and combined comorbidity score, healthcare utilization, number of unique prescribers, number of unique medications, prescriber specialty, pain severity, facility profit status, and number of facility beds. (See Table S1 for covariate definitions and Table S3 for all model results and covariates). Unless otherwise denoted, all comparisons are for presence of that covariate (indicator = yes or 1) vs. no presence of that covariate (indicator = no or 0). ADRD, Alzheimer's disease and related dementias; aRR, adjusted risk ratio; CFS, cognitive function scale.

Potentially modifiable predictors of the prescribing cascade included use of certain medications, including potassium sparing diuretics (aRR = 1.53 [1.10, 2.12]), and thiazide and thiazide‐like diuretics (aRR = 1.27 [1.06, 1.52]). Gabapentinoid type and baseline dose were not significantly associated with the cascade, but receiving 15 or more medications was associated with increased risk (aRR = 1.18 [1.02, 1.37]). However, as shown in Figure 2, the results of the pooled logistic regression model indicated that higher gabapentinoid doses during follow‐up (i.e., post‐baseline) were associated with a statistically significant higher cumulative risk of initiating a loop diuretic. The difference persisted and increased slightly over time. Specifically, 7.2% with a higher dose vs. 5.0% with a lower dose had experienced the prescribing cascade by the end of follow‐up, equivalent to a 1.45‐times higher incidence.

FIGURE 2.

Cumulative incidence of loop diuretics by time‐varying gabapentinoid dose. Lower dose: < 600 mg gabapentin or < 150 mg pregabalin per day. Higher dose: ≥ 600 mg gabapentin or ≥ 150 mg pregabalin per day.

3.3. Stability Analyses

In the stability analyses where we re‐coded all missing outcomes for disenrolled persons, results were largely unchanged (Table S4). Next, when using LASSO regression to choose the best fitting parsimonious model, older age, ADRD diagnosis, level of cognitive impairment, diabetes, number of medications, and use of other types of diuretics were selected as the most important predictors (Table S5). In the less parsimonious but overall best‐fitting model, Lasso selected 32 variables, including calendar year, having a geriatrician as the prescriber, number of unique prescribers, and additional chronic conditions and medication use (Table S6). Performance of the parsimonious and overall best‐fitting models, measured via BIC (−75,602 vs. −75,663, respectively), as well as the aRRs for common variables, was similar.

4. Discussion

In this descriptive observational study, we estimated that approximately 1 in 20 US long‐stay NH residents initiating a gabapentinoid experienced a prescribing cascade involving a loop diuretic within 3 months. Potentially modifiable predictors especially pertinent for NH clinician decision‐making included current receipt of ≥ 15 medications, non‐loop diuretics, and dose increases after initiation. Other predictors associated with increased risk included older age, greater comorbidity, and a diagnosis of neuropathic pain, fibromyalgia, or diabetes. Contrary to expectations, more severe cognitive impairment or a diagnosis of ADRD were associated with lower risk of the gabapentinoid‐loop diuretic cascade. These findings were consistent across multiple stability analyses, including those employing data‐driven variable selection.

Overall, significant predictors in the primary analysis reflected a higher propensity to experience edema. These included age, certain comorbid conditions (e.g., diabetes, higher comorbidity burden), clinical indications for higher doses of gabapentinoids (e.g., neuropathic pain, fibromyalgia), and general indicators of a higher likelihood of being prescribed medications (≥ 15 unique medications). Prior thiazide or potassium‐sparing diuretic use may not be a true modifiable predictor; rather, other diuretic use may reflect a past history of edema and may help to signal to prescribers that a person is likely to experience edema from gabapentinoids. Therefore, the results for prior diuretic use and other potentially modifiable predictors should be interpreted as hypothesis‐generating and potentially examined in future studies testing causal hypotheses with the ultimate goal of guiding interventions. Notably, post‐baseline gabapentinoid dose emerged as an important and modifiable predictor of a gabapentin‐loop diuretic prescribing cascade in our study; however, baseline gabapentinoid dose was not significant. These findings may reflect the fact that older age is associated with known pharmacokinetic and pharmacodynamic changes [41]. For this population of older adults with multimorbidity living in NH, most were started at a lower gabapentinoid dose, yet titration increases common with gabapentinoid use may be more likely to result in edema due to reduced renal function or increased bioavailability [42]. Gradual titration may help to reduce the risk of adverse effects.

In community populations, the incidence of a gabapentinoid‐loop diuretic cascade was substantially lower than in the current cohort (4.2%) of frail, older nursing home residents with multiple comorbidities and high polypharmacy; for example, risk was 0.6% among US adults with employer‐sponsored or Medicare Supplemental insurance in 2005–2019 [27], 1.3% among older US veterans between 2010–2019 [8], and 2.0% among Canadian community‐dwelling older adults with lower back pain in 2011–2019 [26]. Off‐label gabapentinoid use is growing among NH residents. Gabapentinoids are more likely to be initiated among NH residents with pain or diabetes [14]. Similarly, among residents with depression or bipolar disorder but without dementia, gabapentin use as a mood stabilizer increased by 50% between 2011 and 2014 [43]. This growth in off‐label use follows a similar, but lagged, pattern observed in the broader US adult population [44, 45, 46]. As the populations of NH residents receiving gabapentinoids increases, the risk of prescribing cascades is likely to grow as well. Notably, prescribing cascades may be appropriate in certain situations, such as when managing acute, severe edema among patients with high levels of neuropathic pain [9]. Prescribing practices in NHs are also influenced by facility or corporate chain protocols, recommendations from offsite consultant pharmacists, and increasing regulatory pressures related to antipsychotics and opioids [47], all of which may complicate efforts to identify and mitigate inappropriate prescribing cascades.

Several identified predictors were similar to those observed in community‐dwelling older adults, including hyperpolypharmacy [8, 28] and higher gabapentinoid doses [26, 27] as factors associated with increased risk. However, several predictors differed in our population of long‐stay NH residents. Prescriber specialty, including an indicator of the proportion of services delivered in a NH as a measure of a clinician's NH specialization, was not associated with the prescribing cascade. Unlike in prior research [26, 27], we did not observe differences in risk by sex or type of gabapentinoid prescribed. Findings of lower risk in later years could be explained by fewer visits with healthcare providers during the COVID‐19 pandemic or increased awareness among prescribers over time.

Our study has several notable strengths. This is, to our knowledge, the largest population‐based study of the incidence and predictors for the gabapentinoid‐loop diuretic prescribing cascade in long‐stay NH residents to date. We also used a longitudinal, cohort design that offers several advantages over prior prescription sequence symmetry analyses (PSSA) [48, 49]. These include the ability to study all gabapentinoid initiators, not just the subset who initiate both gabapentinoids and loop diuretics selected in PSSA, and to estimate time‐varying measures of absolute risk of the prescribing cascade. Limitations include the nature of administrative claims data, wherein we were unable to reliably determine whether edema occurred after gabapentinoid initiation, the location of the first prescription of the loop diuretic (e.g., ED vs. NH) or identify the true clinical indication for the loop diuretic dispensing. Similarly, due to the nature of Medicare Part D claims, we were required to restrict to a population of stable, long‐stay NH residents with observable medication dispensing data to ascertain incident gabapentin use and subsequent loop diuretic dispensings. Therefore, results may not be generalizable to short‐stay NH residents, such as those receiving post‐acute care. Finally, although we believe the diagnosis of ADRD through claims and time‐varying cognitive function measured by the CFS represent different constructs as covariates, these variables likely have some collinearity. Results identifying lower risk of the prescribing cascade in those with ADRD may need to be interpreted with caution. Despite these limitations, the findings make an important contribution to understanding the incidence and predictors of the gabapentinoid‐loop diuretic prescribing cascade in the vulnerable and understudied long‐stay NH resident population.

5. Conclusions

In this national cohort study of long‐stay NH residents, we found that approximately 1 in 20 residents initiating a gabapentinoid experienced a prescribing cascade within 3 months. Clinicians should monitor residents closely for signs of edema following gabapentinoid initiation and consider dose reduction, deprescribing, or alternative therapies before adding a loop diuretic.

Author Contributions

Kaleen N. Hayes: study concept and design, acquisition of data, analysis and interpretation of data, drafting of the manuscript, critical review of the manuscript for important intellectual content. Emmanuelle Belanger: study concept and design, drafting of the manuscript, critical review of the manuscript for important intellectual content. Arman Oganisian: analysis and interpretation of data, critical review of the manuscript for important intellectual content. Richa Joshi: analysis and interpretation of data, critical review of the manuscript for important intellectual content. Xiao (Joyce) Wang: analysis and interpretation of data, critical review of the manuscript for important intellectual content. Lexie R. Grove: critical review of the manuscript for important intellectual content. Kelsey L. Corcoran: critical review of the manuscript for important intellectual content. Andrew R. Zullo: study concept and design, acquisition of data, critical review of the manuscript for important intellectual content.

Funding

The work submitted in this manuscript was sponsored by the National Institute on Aging, award number (RF1AG089541).

Disclosure

K.N.H. has received grant funding paid directly to Brown University for investigator‐initiated research from Sanofi, Genentech, and GlaxoSmithKline for research on influenza vaccination in nursing homes, influenza outbreak control, and shingles vaccination in nursing homes, respectively. A.R.Z. has received grant funding paid directly to Brown University by Sanofi for collaborative research on the epidemiology of infections and vaccinations among nursing home residents and infants. E.B., A.O., R.J., X.W., L.R.G., K.L.C. report no potential conflicts of interest.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Exclusion criteria and covariate definitions.

Table S2: All measured characteristics of nursing home residents who initiated gabapentinoids.

Table S3: Univariate and adjusted relative risk ratios (RRs) for predictors of the gabapentinoid‐

Table S4: Results from stability analyses recoding all missing outcome values to “yes” for

Table S5: Variables selected by LASSO methods (selection based on Bayesian Information

Table S6: Variables selected by LASSO methods (selection based on Bayesian Information

Figure S1: Study diagram.

Figure S2: Exclusion flow diagram.

Hayes K. N., Belanger E., Oganisian A., et al., “Predictors of a Gabapentinoid–Loop‐Diuretic Prescribing Cascade in U.S. Nursing Home Residents,” Journal of the American Geriatrics Society 74, no. 2 (2026): 336–344, 10.1111/jgs.70219.