The Association Between the Number of Prescription Medications and Incident Falls in a Multi-ethnic Population of Adult Type-2 Diabetes Patients: The Diabetes and Aging Study

Abstract

Background

Use of four or more prescription medications is considered a risk factor for falls in older people. It is unclear whether this polypharmacy-fall relationship differs for adults with diabetes.

Objective

We evaluated the association between number of prescription medications and incident falls in a multi-ethnic population of type-2 diabetes patients in order to establish an evidence-based medication threshold for fall risk in diabetes.

Design

Baseline survey (1994-1997) with 5 years of longitudinal follow-up.

Participants

Eligible subjects (N = 46,946) had type-2 diabetes, were ≥18 years old, and enrolled in the Kaiser Permanente Northern California Diabetes Registry.

Measurements and main results

We identified clinically recognized incident falls based on diagnostic codes (ICD-9 codes: E880-E888). Relative to regimens of 0-1 medications, regimens including 4 or more prescription medications were significantly associated with an increased risk of falls [4-5 medications adjusted HR 1.22 (1.04, 1.43), 6-7 medications 1.33 (1.12, 1.58), >7 medications 1.59 (1.34, 1.89)]. None of the individual glucose-lowering medications was found to be significantly associated with a higher risk of falls in predictive models.

Conclusions

The prescription of four or more medications was associated with an increased risk of falls among adult diabetes patients, while no specific glucose-lowering agent was linked to increased risk. Baseline risk of falls and number of baseline medications are additional factors to consider when deciding whether to intensify diabetes treatments.

INTRODUCTION

The focus of diabetes care has traditionally been on the prevention of microvascular and cardiovascular complications. Clinical trials have demonstrated that the risks for these complications can be reduced with comprehensive control of glucose, blood pressure, and cholesterol levels.^1,2 These trials have provided the supporting evidence for population-level risk factor target goals in clinical practice guidelines. The current quality of diabetes care in the population indicates that many patients will require more intensive use of medications to achieve these goals.^3,4

While preventing the traditional complications of diabetes is of clear public health importance, there is increasing recognition that diabetes and its treatment are also associated with an increased risk of non-traditional complications such as falls, hip fractures, incontinence, chronic pain, and depression, especially among elderly, frail patients.⁵ In recent years, the American Diabetes Association⁶ and the American Geriatrics Society⁵ have formally recommended that providers devote greater attention to preventing and treating these non-traditional complications.

Polypharmacy, the use of multiple prescription medications, is the non-traditional complication that probably has the most direct implications for diabetes management.⁷ The use of four or more prescription medications is considered an established risk factor for falls in older people.⁸ This clinical rule of thumb is based on observational studies demonstrating a relationship between the number of prescription drugs and falls^9–12 or impaired balance¹³, and on clinical trials of fall reduction interventions that have included an effort to actively reduce the use of specific psychoactive drugs and/or the overall number of medications.^14–16

Despite its clear clinical importance, the association of prescription medications and falls has not been frequently studied in diabetes. There are multiple reasons to reexamine this relationship in the specific context of diabetes. First, the relationship between prescription medications and falls in diabetes patients may be confounded by diabetes-specific variables such as diabetes duration and glucose control.¹⁷ Second, it is important to establish a new evidence-based medication threshold for fall risk in diabetes, given that the average number of diabetes-related medications that patients are prescribed has already reached the traditional fall risk threshold of four medications.³ Third, individual glucose-lowering medications may be independent risk factors for falls.¹⁸ Fourth, the relationship between prescription medications and falls has never been examined in a multi-ethnic cohort or among younger patients with chronic diseases such as diabetes. To address these issues, we evaluated the association between prescribed medications and fall rates in a multi-ethnic population of adult diabetes patients with uniform access to care.

METHODS

Source Population

The Kaiser Permanente Northern California Diabetes Registry (the “registry”) is a well-characterized population that has been maintained continuously since 1993.^19–21 Registry eligibility is based on multiple sources of data including pharmacy (diabetes medication prescriptions), laboratory (A1C ≥7%), and outpatient, emergency room and hospitalization diagnoses of diabetes. The registry was 99.5% sensitive for diagnosed diabetes, compared with self-report, as of January 2003. All automated clinical information (pharmacy, laboratory, outpatient and inpatient diagnoses and procedures) is downloaded annually to provide a comprehensive, longitudinal follow-up of each registry member.

To collect data not captured electronically, all members of the registry identified prior to January 1, 1996 and 18 years or older (n = 136,537) were surveyed by mail and computer-assisted telephone interview in 1994-1997 in English or Spanish. After removing non-responders and those not asked questions on height and weight, 65,859 individuals remained. We further excluded patients who had died before the baseline date (998), with less than 12 months of continuous Kaiser membership before baseline (2,990), less than 12 months of drug benefit before baseline (6,018), prevalent falls within 5 years prior to baseline (865), and unknown type of diabetes or type-1 diabetes (8,042). Type-1 diabetes was defined by clinical characteristics using a previously published algorithm.²¹ The remaining 46,946.subjects were the basis for this analysis.

Timeframe for Analysis

The baseline for analysis was defined as the date of completion of the 1994-1997 survey or January 1, 1996, whichever was later. We started follow-up after January 1, 1996 because March 1, 1995 was the date when the prescription drug database was fully implemented across all facilities, providing us with at least 10 months of complete, pre-baseline drug utilization information. End of follow-up for the analysis was defined as time at the first fall (endpoint of interest), death, end of Kaiser membership, or end of drug benefit (each defined as a gap of at least 3 months in coverage), or 5 years after baseline.

Incident Falls

Incident falls were identified from inpatient and outpatient diagnostic codes (ICD-9 codes: E880-E888) over 5 years of follow-up. We also identified whether each incident fall was associated with a fracture (ICD-9 codes: 733, 800-829, E887) on the same day.

Defining Prescribed Medications

The main exposure of interest was the total number of prescribed medications at baseline. Because there is no commonly accepted definition of medication use or polypharmacy in the literature⁷, we developed our own protocol for identifying and counting prescriped medications. When counting medications, we required that medications have the potential for long-term use and that they represent distinct pharmacological agents. We defined a prescribed medication as one filled prescription for a given medication, for at least a 30 days’ supply, within the 6 months prior to baseline. Over-the-counter medications that might be prescribed for acute illnesses were excluded with the exception of antihistamines, aspirin, calcium, and magnesium. We included chemotherapeutic agents and contraception that might be delivered during clinical visits. For combination medications, we counted each distinct pharmacological agent as a prescription medication. Different doses of the same medication that might be taken at different times of the day were counted as one medication. For patients prescribed multiple medications from the same general class of medications (e.g., diuretics), we counted each distinct medication within the class as an individual drug. Two physicians reviewed all prescribed medications to determine if they met these general criteria (Drs. Huang and Ahmed); differences of opinion were reconciled during regular meetings.

We considered the total number of prescribed medications as indicator variables for different cutpoints of total medications. The associations between the prescription of individual glucose-lowering medications and incident falls were considered separately. At the time of the survey, the primary means of treating diabetes were insulin, sulfonylureas, and metformin.

Assessment of Covariates

The major baseline covariates evaluated in this analysis were demographics (age, gender, race/ethnicity), body composition (body mass index), and health behaviors (self-reported smoking and alcohol consumption²²). We also included variables related to the state of diabetes including time since diabetes diagnosis based on self-report, glycosylated hemoglobin (HbA_1C) within 1 year prior to baseline from clinical laboratory data, hypertension based on self-report and/or prescription of antihypertensives, laser photocoagulation from January 1994 to baseline survey from administrative data, peripheral neuropathy based on self-reported symptoms, end-stage renal disease based on current dialysis or history of transplantation from administrative data, lower extremity amputation based on any prior administrative history, myocardial infarction based on past 5-year administrative history, congestive heart failure based on any prior administrative history, stroke based on any prior administrative history, and number of hospitalizations in the preceding 1 year.^21,23

In our in-depth analysis of the predictive effect of individual glucose-lowering medications, we also adjusted for the prescription of other medications that are known for to be associated with falls [beta blockers, diuretics, antiarrhythmics (class 1a), digoxin, benzodiazepindes, neuroleptics, antidepressants, and antiseizure medications].⁸

Statistical Analysis

All analyses were performed with SAS version 9.1 (SAS Institute, Cary, NC), and associations were considered statistically significant at p < 0.05. We first calculated crude and age-sex adjusted rates (incidence densities) of falls (events/1,000 person years) for the entire cohort and for major racial/ethnic groups.

We used Cox proportional hazards models to evaluate the association of the number of prescription medications and individual glucose-lowering medications with incident falls. Univariate analyses were first performed followed by multivariate analysis accounting for all previously described covariates. The total prescription medication models did not specify individual glucose-lowering medications as covariates. For our evaluation of the role of individual glucose-lowering medications, we used models intended to be predictive models of falls. Because these analyses were intended to identify individual medications that contributed most to fall risk, we did account for individual classes of medications classically associated with falls. We evaluated interaction terms of prescribed medication with race, sex, and age in models and performed stratified analysis when the interaction term was significant.

Sensitivity Analyses

We conducted a sensitivity analysis using an alternative definition of a long-term prescribed medication based on a baseline 30-day supply, followed by at least one refill within 365 days of the initial fill. We also evaluated the dynamic nature of prescription medication over time using an extended Cox model.

RESULTS

Table 1 characterizes our study sample. The population was racially diverse. The majority of patients had been diagnosed with diabetes for less than 10 years at baseline and had a mean HbA_1C of 8.34%. The mean total number of prescribed medications was 4.16 (SD 3.27). The most frequently prescribed medications for glucose control were sulfonylureas followed by any form of insulin and metformin.

Table 1.

Characteristics of Adult Patients with Type-2 Diabetes by Incident Fall Status

	Overall (N = 46,946)	Incident fall (N = 1,807)	No incident fall (N = 45,139)	P-values*
Demographics
Mean age (SD)	61.56 (12.16)	69.87 (10.77)	61.22 (12.10)	<0.01
Age category (%)
18-49	18.07	5.42	18.58	<0.01
50-64	37.61	20.09	38.31
65-74	29.36	37.02	29.05
75-84	13.23	32.10	12.48
≥85	1.73	5.37	1.58
Female (%)	46.43	57.83	45.97	<0.01
Ethnicity (%)
White	54.30	64.80	53.88	<0.01
African American	10.57	6.75	10.73
Latino	5.27	3.15	5.36
Asian	7.78	4.10	7.93
Other	21.96	21.20	21.99
Education (%):
High school or less	44.03	51.25	43.74	<0.01
Some college	30.53	27.17	30.67
College graduate	23.99	20.20	24.15
Clinical characteristics
BMI (%)
<18.5	0.56	1.33	0.53	<0.01
18.5-24.9	18.38	25.57	18.10
25-29.9	35.55	35.03	35.57
30-34.9	23.84	21.08	23.95
≥35	18.38	12.01	18.63
Missing	3.30	4.98	3.23
A1C, mean (SD)	8.34 (1.85)	8.29 (1.84)	8.34 (1.85)	0.39
A1C (%)
No test available	32.37	34.53	32.29	0.33
<6	4.53	4.65	4.52
6-6.9	14.28	13.89	14.30
7-7.9	15.72	15.66	15.72
8-9.9	20.97	20.42	20.99
≥10	12.13	10.85	12.18
Duration of diabetes (%)
<10 years	61.22	44.88	61.87	<0.01
10-19 years	21.93	27.89	21.69
≥20 years	11.11	19.59	10.77
Missing	5.75	7.64	5.68
Hypertension (%)	54.36	58.16	54.21	<0.01
Laser photocoagulation (%)	2.97	4.54	2.91	<0.01
Peripheral neuropathy (%)	24.88	32.76	24.57	<0.01
Lower extremity amputation (%)	0.99	1.44	0.97	0.05
End-stage renal disease (%)	2.69	4.48	2.62	<0.01
Myocardial infarction prior 5 years (%)	6.76	9.30	6.66	<0.01
Congestive heart failure (%)	6.44	11.68	6.23	<0.01
Stroke (%)	3.43	7.03	3.28	<0.01
Number of hospitalizations 1 year prior to baseline, mean (SD)	0.19 (0.60)	0.28 (0.66)	0.18 (0.59)	<0.01
Medications
Mean total number of medications (SD)	4.16 (3.27)	5.16 (3.63)	4.12 (3.25)	<0.01
Total number of medications (%)
0	7.87	5.48	7.97	<0.01
1	13.47	8.74	13.66
2	15.02	10.96	15.18
3	13.75	11.62	13.83
4	12.22	12.78	12.20
5	9.91	11.34	9.85
6	7.81	8.74	7.77
7	5.80	8.30	5.70
>7	14.15	22.04	13.84
Antiarrhythmics (%)	0.72	1.33	0.69	<0.01
Digoxin (%)	6.59	12.67	6.34	<0.01
Benzodiazepine (%)	5.48	8.47	5.36	<0.01
Neuroleptic (%)	0.99	0.83	1.00	0.47
Antidepressant (%)	12.02	17.32	11.81	<0.01
Antiseizure (%)	0.93	1.66	0.90	<0.01
Mean number of diabetes-related medications (SD)	2.05 (1.47)	2.32 (1.53)	2.04 (1.46)	<0.01
Glucose-lowering medication use (%)
Insulin	23.16	28.61	22.94	<0.01
Sulfonylurea	58.47	56.83	58.54	0.15
Metformin	5.11	5.09	5.12	0.96
Blood pressure medication use (%)
Beta-blocker	13.26	14.61	13.21	0.08
Diuretics	24.61	36.52	24.13	<0.01
ACE inhibitor	27.96	30.77	27.85	<0.01
Alpha-blocker	4.98	5.87	4.95	0.08
Calcium channel blocker	20.07	21.97	19.99	0.04
Centrally acting antihypertensive	3.80	5.70	3.72	<0.01
Vasodilator	1.25	1.66	1.23	0.11
Self-care behaviors
Smoking (%)
Never	46.86	47.32	46.84	<0.01
Former	40.70	40.68	40.70
Current	11.03	9.57	11.09
Heavy drinking (%)				<0.01
Yes	15.22	13.67	15.28
No	79.98	79.14	80.01
Missing	4.80	7.19	4.71

*P-values calculated from chi-square (categorical) and Student's t-tests (comparison of means)

Diabetes-related medications were defined as all glucose-lowering drugs, cholesterol-lowering drugs, blood pressure-lowering drugs, and aspirin

The mean follow-up time to incident event or censoring was 4.18 years (SD 1.44 years). Four percent of patients experienced an incident fall. Fifty-two percent of incident falls were associated with a fracture on the day of diagnosis. The overall crude incidence of falls was 9.21/1,000 person years (py). Incidence density differed by age, gender, and race/ethnicity. The incidence density steadily rose with advancing age, ranging from 2.74/1,000 py for patients 18-49 years of age to 11.47/1,000 py for patients 65-74 years of age. Women were more likely to fall than men [age-adjusted rate for women 11.40/1,000 py, men 7.27/1,000 py (p-value <0.01)]. Among the racial/ethnic groups, Non-Hispanic whites had the highest incidence of falls after adjustment for age and sex (10.16/1,000 py), followed by other race (9.27/1,000 py), Latinos (7.48/1,000 py), Asians (6.53/1,000 py), and African Americans (5.84/1,000 py) (all minority-white comparisons, p-values <0.01). Non-Hispanic whites had the highest adjusted incidence of falls associated with fractures; all ethnic groups had nearly identical rates of falls not associated with fractures.

Apart from differences by race/ethnicity, patients with falls had other notable differences from those who did not fall (Table 1). Patients who fell were older, less educated, more likely to have longer duration of diabetes, more likely to have diabetes-related complications, and more likely to take cardiovascular and psychotropic medications associated with falls. There was no difference in the mean or distribution of baseline glucose control by fall status.

Individuals who experienced falls were prescribed a larger number of medications than those who did not (Table 1). We found no significant increase in risk of falls with the prescription of up to three baseline medications; however, we found a significant and monotonic increase in the risk of falls with the prescription of four or more baseline medications after adjustment, especially for those under age 65 (Table 2). In sensitivity analyses, using the alternative definition of total medications reduced the mean number of medications from 4.2 to 3.1 medications. We found that patients with regimens of 4 or more medications had a statistically higher risk of falls than patients with 0-1 medications. When we evaluated the total number of medications using a Cox model with time-varying exposure variables, the relationship between total medications and falls strengthened.

Table 2.

Hazard Ratios (95% Confidence Intervals) for the Total Number of Baseline Prescription Medications and Incident Falls Among Patients with Diabetes

	Overall population	<65 years of age	≥65 years of age
2-3 medication	1.04 (0.89, 1.22)	1.14 (0.84, 1.55)	1.00 (0.83, 1.20)
4-5 medications	1.22 (1.04, 1.43)	1.45 (1.06, 1.99)	1.13 (0.94, 1.35)
6-7 medications	1.33 (1.12, 1.58)	1.34 (0.93, 1.92)	1.28 (1.05, 1.56)
>7 medications	1.59 (1.34, 1.89)	2.31 (1.65, 3.23)	1.39 (1.14, 1.70)

All models adjusted for gender, age, race, smoking, alcohol consumption, body mass index, duration of diabetes, hemoglobin A1C, neuropathy, lower extremity amputation, end-stage renal disease, myocardial infarction, congestive heart failure, stroke, laser photocoagulation, and prior hospitalizations. The referent group is 0-1 prescription medications

In our analysis of individual glucose-lowering medications, the prescription of insulin was found to confer an increased risk of falls in unadjusted analyses [HR 1.39 (95% CI 1.25, 1.53), referent category no insulin] (Table 3). However, after adjusting for other medications and clinical variables, this association became non-significant. The other classes of glucose-lowering medications did not have any statistically significant association with falls.

Table 3.

Hazard Ratios (95% Confidence Intervals) for Individual Glucose-Lowering Agents and Incident Falls

	Model 1	Model 2
Insulin	1.27 (1.15, 1.41)	1.11 (0.99, 1.24)
Sulfonylurea	0.96 (0.88, 1.06)	1.04 (0.95, 1.15)
Metformin	1.17 (0.95, 1.45)	1.16 (0.94, 1.44)

Model 1: Adjusted for age, gender, race, and other medications (beta-blockers, diuretics, anti-arrhythmics, digoxin, benzodiazepines, neuroleptics, anti-depressants, anti-seizure medications, ACE inhibitors, alpha-blockers, calcium channel blockers, centrally-acting anti-hypertensives, vasodilators, and estrogen)

Model 2: Adjusted for age, gender, race, other medications (see model 3), smoking, alcohol consumption, body mass index, duration of diabetes, hemoglobin A1C, neuropathy, lower extremity amputation, end-stage renal disease, myocardial infarction, congestive heart failure, stroke, laser photocoagulation, and prior hospitalizations

We also specified models with cross-product terms to evaluate whether the relationship between medications and falls differed by race/ethnicity and age. We did not find statistically significant interactions between ethnicity and prescription of specific glucose-lowering drugs or between ethnicity and total number of medications. In our analysis examining the relationship between total number of medications (as indicator variables) and falls by different age groups (<65, ≥65 years of age), we found a statistically significant overall interaction (p < 0.01). In age-stratified analyses, we found that the hazard ratios and confidence intervals for total number of medications were not substantively different across major age groups. In our analysis of total medications as indicator variables in younger patients, we found a significant increase in the risk of falls with 4-5 and 7 or more baseline medications, but a non-significant increase in risk with 6-7 medications. In older patients, there was a monotonic rise in the risk of falls that became statistically significant at 6 or more baseline medications (Table 2).

DISCUSSION

Our study confirms that an increasing number of prescription medications is independently associated with the risk of falls in a multi-ethnic sample of community dwelling adults with type-2 diabetes. This association remained significant despite extensive adjustment for diabetes-related variables and health behaviors. Consistent with established practice guidelines for older patients, we found that fall risk increases steadily at four or more prescription medications in this diabetes population.⁸ Patients who were prescribed four or more baseline medications had an incidence density for falls of 11.94/1,000 py compared to an incidence density of 6.61/1,000 py for those who were prescribed fewer medications. We did not find that the prescription of any individual glucose-lowering medications conferred an additional increased risk of falls.

Reexamining the association between prescription medication and falls in diabetes is important given that we currently face conflicting challenges in diabetes care. A major public health message in diabetes care is that overall medication management of the population needs to be intensified given observed suboptimal levels of clinical control. However, recent data from the Action to Control Cardiovascular Risk in Diabetes showing that aggressive use of glucose-lowering medications increased the risk of mortality raise questions about the potential harms of intensive pharmacotherapeutic management of diabetes.²⁴ Our study raises another concern by suggesting that fall risk will likely be exacerbated by an intensified use of prescription medications needed to achieve tight control.

In comparison to prior studies of medications and falls, our study is distinctive due to its relatively large sample size (~47,000 vs. 4,108 for the next largest study²⁵), the use of pharmaceutical utilization records instead of medical records review or patient interviews for counting medications⁹, and the use of diagnostic codes to establish the incidence of falls instead of chart review or patient interviews.¹⁰ These differences most likely had competing effects on the strength of association between medications and falls. Our large sample size conferred greater statistical power, whereas our fall ascertainment method most likely undercounted events compared to past studies. Despite these differences, this study reconfirmed the presence of the relationship of medications and falls in the elderly with diabetes.

Perhaps the most important differences between our study and past work relate to the ethnic and age diversity of the study population. In our multi-ethnic population, we found that fall risk was highest in non-Hispanic whites. This excess risk is likely attributable to higher rates of osteoporosis among whites.²⁶ We found that whites had the highest incidence of falls associated with fractures, but not for falls not associated with fractures. This suggests that falls in whites were probably coming to medical attention more frequently given their higher likelihood of osteoporosis-related fractures. While there were ethnic differences in the risk of falls, we did not find clinically significant ethnic differences in the association between medications prescribed and the risk of falls. In our comparison by age groups, we found that older patients had a higher fall risk than younger patients; however, the medication-fall association was generally similar in both age groups. In fact, the medication threshold at which fall risk begins to rise may be lower in younger patients than older patients. The fact that a medication-fall association exists among community-dwelling younger patients with diabetes is a novel observation.

There are several limitations of this study. There is no commonly accepted approach to determining the number of prescription medications that a patient is taking. With our particular pharmaceutical data and protocol, we undercounted over-the-counter medications such as aspirin and included medications that might be excluded by others and may have overcounted chronic medications. In sensitivity analyses, we found that the relationship between drug count and falls was robust despite using alternative definitions of total number of medications. Apart from this, the original survey of the Kaiser Permanente Northern California Diabetes Registry took place in the mid-to-late 1990s, and diabetes care has changed significantly since that time. The availability of newer diabetes-related agents and the removal of older medications may alter the association between the number of prescription medications and falls and will require reanalysis with more contemporary data once sufficient data have accrued. In addition, our identification of falls was based on a diagnostic code algorithm; this definition likely identified clinically recognized falls but undercounted minor falls. Future research of the association between medication and falls would benefit from formal validity studies of diagnostic protocols for falls. Our cohort was also limited to patients enrolled in a managed care system, and diabetes care in such systems may be systematically different from that in other clinical settings; however, it is unlikely that fundamental associations between prescribed medications and fall events would differ across clinical settings. Finally, while we made every effort to adjust for possible known confounders of the association between medications and falls, there may still be unobserved confounding.

Despite these limitations, our study confirms an independent association between the total number of medications that diabetes patients are prescribed and incident falls. This finding suggests that a sizeable proportion of patients living with diabetes are subjected to an increased risk of falls given that 50% of our patients were prescribed four or more medications and likely used at least three or more chronically. Future research examining the relationship between medications and adverse events such as falls would benefit from more formal evaluations of what defines polypharmacy and what mechanisms link polypharmacy to adverse events.