Incident Dementia, Glycated Hemoglobin (HbA1c) Levels, and Potentially Preventable Hospitalizations in People Aged 65 and Older With Diabetes

Abstract

Background

This study was aimed to determine whether incident dementia and HbA1c levels are associated with increased rates of potentially preventable hospitalizations (PPHs) in persons with diabetes.

Method

A total of 565 adults aged 65+ ever treated for diabetes were enrolled from Adult Changes in Thought study. PPHs were from principal discharge diagnoses and included diabetes PPH (dPPH), respiratory PPH (rPPH), urinovolemic PPH (uPPH), cardiovascular PPH, and other PPH. Poisson generalized estimating equations estimated rate ratios (RRs) and 95% confidence intervals (CIs) for the associations between dementia or HbA1c measures and rate of PPHs.

Results

A total of 562 individuals contributed 3 602 dementia-free years, and 132 individuals contributed 511 dementia follow-up years. One hundred twenty-eight (23%) dementia-free individuals had 210 PPH admissions and a crude rate of 58 per 1 000 person-years, while 55 (42%) individuals with dementia had 93 PPH admissions and a crude rate of 182 per 1 000 person-years. The adjusted RR (95% CI) comparing rates between dementia and dementia-free groups were 2.27 (1.60, 3.21) for overall PPH; 5.90 (2.70, 12.88) for dPPH; 5.17 (2.49, 10.73) for uPPH; and 2.01 (1.06, 3.83) for rPPH. Compared with HbA1c of 7%–8% and adjusted for dementia, the RR (95% CI) for overall PPH was 1.43 (1.00, 2.06) for >8% HbA1c and 1.18 (0.85, 1.65) for <7% HbA1c. The uPPH RR was also increased, comparing >8% and <7% HbA1c levels.

Conclusion

Incident dementia is associated with higher rates of PPHs among people with diabetes, especially PPHs due to diabetes, urinary tract infection (UTI), and dehydration. Potential evidence suggested that HbA1c levels of >8% versus lower levels are associated with higher rates of overall PPHs and UTI- and dehydration-related PPHs.

As people worldwide live longer, persons with both dementia and type 2 diabetes (hereafter diabetes) are increasingly prevalent in primary and ambulatory care (1). Diabetes is a chronic condition that should be treated and managed in a community. Persons with dementia and diabetes, however, have higher risks of hospitalization due to diabetes and other complications, and many of these hospitalizations are considered preventable and potentially avoidable (2–4). According to the Agency for Healthcare Research and Quality (AHRQ), potentially preventable hospitalizations (PPHs) due to ambulatory care sensitive conditions are “hospitalizations for health problems that could have been avoided with adequate primary care or disease management in outpatient settings” (5). Previous Medicare data showed almost a 4-fold higher risk of PPHs related to diabetes in persons with dementia (2), with such hospitalizations being expensive and harmful in terms of increased risk of delirium, cognitive decline, and iatrogenesis (6,7). These national claim-based estimates, however, should be updated to include other common reasons for PPHs and strengthened through a more extensive adjustment for comorbidities and earlier dementia diagnosis by utilizing higher-quality data that integrate both administrative and clinical sources and implement proactive dementia identification protocols.

Another important consideration in a person with diabetes is that both the risk of diabetes complications and the risk of treatment should be considered when setting therapeutic goals. Poor glycemic control over time may lead to significant osmotic symptoms (eg, diuresis) if the renal threshold for glycosuria is crossed, as well as ketoacidosis and a hyperosmolar hyperglycemic state. These changes, in turn, can lead to dehydration, dizziness, falls, infections, accelerated cognitive decline, and hospitalizations (8). Conversely, tight glycemic control might also result in more frequent hypoglycemia with vascular and neurological compromise in the brain, falls, and utilization of acute care (9). Thus, both tight and poor glycemic control can lead to hospitalizations due to complications, but the most appropriate glycemic targets for patients with dementia have not been well defined. The American Diabetes Association (ADA) has promulgated recommendations on optimal glucose targets in “complex” and “very complex” patients with cognitive impairment (10), but these consensus statements should be strengthened with higher levels of evidence. The exclusion of persons with dementia from most of the randomized controlled trials of diabetes interventions has left a void in the diabetes literature regarding a population with high prevalence rates and complex health needs. Although observational studies can provide some useful indirect evidence on glycemic goals in this population (11,12), methodological weaknesses such as sparsely collected glucose measures and retrospective capture of dementia diagnosis have precluded conclusive insights into the “real-world” glucose management.

We used a unique longitudinal study of aging and dementia to offset the above-mentioned limitations to determine the extent to which dementia, as diagnosed prospectively using a research-based protocol, is associated with higher rates of different PPH admissions among persons with diabetes. We also evaluated the association of glucose levels, in terms of HbA1c measurements collected during regular clinical encounters, with PPH rates in persons with diabetes before and after a dementia diagnosis.

Method

Setting

Adult Changes in Thought (ACT) is a prospective cohort study to understand risk factors for the development of dementia among older adults (13). ACT enrollment and follow-up began in 1994–1996 and has been ongoing, with new participants periodically added to replenish active participant numbers. The ACT study randomly samples and enrolls members from Kaiser Permanente Washington (KPWA), an integrated health care delivery system in the state of Washington, who are at least 65 years of age and without dementia, and then follows them with biennial study visits. At those visits, study staff collect information on numerous participant health characteristics and screen them for dementia. Additionally, information on health care utilization, including prescription fills, laboratory measures, clinical encounters, and diagnoses, are available on participants from their extensive electronic health records (EHRs) maintained by KPWA. Study procedures were approved by institutional review boards of KPWA and the University of Washington, and participants provided written informed consent.

Participants

Our analyses were limited to ACT participants who were treated for diabetes, which was defined as having at least 2 fills of antidiabetic medication in a 12-month period. Diabetes onset was defined on the second fill date. The KPWA electronic data on medication refills goes as far back as 1977. We further restricted to participants who did not have dementia diagnosis before diabetes onset, enrolled in KPWA for at least 2 years before entering the study cohort, and had at least 2 HbA1c measures in those 2 years. Participants were required to have at least one ACT follow-up visit (to capture dementia) and to be enrolled in KPWA at the time of follow-up visits (to ascertain hospitalization outcomes). Participants entered the study cohort (hereafter referred to as “baseline”) at the first ACT visit following diabetes onset. They were followed until the earliest of death, withdrawal from ACT, disenrollment from KPWA, September 30, 2018 when data collection ended, and for individuals who did not develop dementia during study follow-up, 1 year following the most recent biennial ACT visit. Per ACT protocol, participants would not return for biennial visits once they had a visit resulting in a dementia diagnosis but were followed remotely through phone and medical records review. Thus, in our analysis, they are followed after their dementia diagnosis as long as they were enrolled in KPWA where we could capture their laboratory and hospitalization data.

Exposure

At ACT biennial visits, participants were screened for dementia using the Cognitive Abilities Screening Instrument (CASI) (14). The CASI ranges from 0 to 100, with higher scores reflecting better cognition. Those who scored less than 86 on the CASI received an in-depth diagnostic workup, including a clinical and cognitive evaluation, which was reviewed by a multidisciplinary committee that assigned dementia diagnoses according to Diagnostic and Statistical Manual of Mental Disorders, 4th ed. (SM-IV) criteria (13). Dementia onset was defined as the mid-point between the ACT visit when dementia was diagnosed and the previous ACT visit (13).

Clinical HbA1c measures were captured from the KPWA computerized laboratory data. As in a previous study (15), we calculated average HbA1c levels in 2-year rolling windows (defined in “Statistical Analyses” section), which roughly approximated time periods between the ACT study visits. A small percentage of participants (4.2%) had at least one 2-year period that had no HbA1c measures. We imputed the missing value by averaging all previous values taken on the same person since the 2 years before baseline. The average levels were then categorized into 3 groups (<7%, 7%–8%, and >8%).

Outcomes

All-cause hospital admissions requiring an overnight stay were collected from participants’ EHR at KPWA. PPHs by type were classified using the principal discharge diagnosis according to AHRQ and other sources (5,16,17). Specifically, we identified overall PPH based on International Classification of Diseases-9th/10th revision diagnosis codes for the conditions included in Supplementary Table S1. Diabetes PPHs (dPPHs) were based on diabetes with short- or long-term complications, low-extremity amputation, diabetes without mention of complication, and hypoglycemia diagnoses. Respiratory PPHs (rPPHs) were based on pneumonia, influenza, and asthma/chronic obstructive pulmonary disease diagnoses. Urinovolemic PPHs (uPPHs) were based on urinary tract infection (UTI) and dehydration diagnoses. Cardiovascular PPHs (cPPHs) were based on congestive heart failure and hypertension angina diagnoses. Other acute PPHs (oPPHs) were based on appendix, cellulitis, gastric/duodenal/peptic ulcer, gastroenteritis, and seizure diagnoses. Similar to HbA1c levels, the number of hospitalizations (all-cause, overall PPH, and PPH subtypes) were summarized in each 2-year rolling window.

Other Covariates

Covariates collected at baseline included sex, self-reported race/ethnicity (non-Hispanic White vs other), education (at least some college vs high school or less), duration of diabetes (measured since earliest evidence of treatment), and whether participants live alone. Having at least one difficulty in instrumental activities of daily living (IADL) and at least 3 in basic activities of daily living (ADL) were computed at baseline based on ACT questionnaires. Frailty was defined using a previously described method (18). Charlson comorbidity score that excluded diabetes diagnosis and inpatient stays (to avoid including previous hospitalization outcomes) was computed each year based on diagnoses recorded in the EHR. Diabetes medications in 2 years prior to the study baseline were from pharmacy fills and included the following mutually exclusive groups: insulin only, metformin only, sulfonylureas only, and mixed. Nursing home admissions during the study follow-up were collected from the EHR.

Statistical Analyses

We divided the study follow-up time for each person into multiple 2-year intervals. The average HbA1c level in the 2 years prior to the current interval was used as exposure of interest in order to eliminate any effects on HbA1c from hospitalizations (outcomes) that occurred in the current interval. The 2-year interval was further divided into times before and after dementia diagnosis if the diagnosis occurred during the interval. People who developed dementia contributed time at risk to both the nondementia and dementia groups.

We compared the crude rates of all-cause hospitalizations, PPHs, and each type of PPH before and after a dementia diagnosis, as well as between the 3 HbA1c level groups. Crude hospitalization rates reported per 1 000 person-years were calculated as the total number of hospitalizations in each group divided by the total person-years of follow-up in that group.

We compared the ratios of hospitalization rates between exposure groups using Poisson regressions with a rate ratio (RR) and 95% confidence interval (CI) estimated from generalized estimation equations with an independence working correlation matrix to account for the correlations of multiple intervals for the same person, and empirical standard errors to account for overdispersion in Poisson regression models. We conducted 3 models for each outcome: (i) crude or unadjusted, (ii) adjusted for age and sex, and (iii) adjusted for age, sex, race, education, living arrangement, IADL, duration of diabetes, and Charlson comorbidity score. Charlson score and age were included as time-varying covariates, with Charlson being assessed in a year prior to each 2-year interval and age being modeled using a restricted cubic spline with knots at tertiles. All other covariates were defined at baseline.

To better understand to what extent events such as impending death and nursing home placement among persons with dementia could alter patterns of medical care and influence decisions regarding hospitalization, we performed 2 sensitivity analyses for the association between dementia diagnosis and rates of hospitalizations. First, we excluded the 2-year follow-up period in which a person died for the 187 participants who died during study follow-up and repeated the multivariate analyses. Second, we further adjusted for a time-varying measure for nursing home placement in the fully adjusted models. We carried forward the participants’ first nursing home placement through the end of the study follow-up.

To understand the extent to which frailty, which is a geriatric condition with glucoregulation pathology (19), affected the association between HbA1c level and rates of hospitalizations, we included interaction terms between frailty defined at baseline and HbA1c level in the fully adjusted models. All analyses were performed in SAS version 9.4 (SAS Institute, Cary, NC).

Results

Participant Characteristics

A total of 565 ACT participants met eligibility criteria and were included in the analyses (Figure 1). Among 132 participants who developed dementia, 8 (6%) disenrolled from KPWA or withdrew from ACT, and 103 (78%) died before September 30, 2018. Among the remaining 433 participants who were dementia-free at their most recent ACT study visit, 49 (11%) left KPWA or ACT, and 84 (19%) died before September 30, 2018.

Figure 1.

Participant flow, vital status, and person-years at the end of follow-up. KPWA = Kaiser Permanente Washington.

Table 1 shows participant characteristics at the study baseline by average HbA1c levels (<7%, 7%–8%, and >8%) in the 2 years prior to cohort entry. Differences across HbA1c groups were observed for race/ethnicity, living alone, and duration of diabetes. There were also differences in diabetes management according to the HbA1c levels. Specifically, people in the poor glycemic control group (>8% HbA1c) were more likely to be treated with a combination of insulin and oral antidiabetic medications than people in lower HbA1c groups.

Table 1.

Characteristics of the Study Participants at Baseline Overall and by Average HbA1c Levels in 2 Years Prior to Study Baseline

	Total (N = 565)		HbA1c Level in 2 y Prior to Study Baseline
			<7% (n = 264)		7%–8% (n = 190)		>8% (n = 111)
Characteristics	n	(%)a	n	(%)a	n	(%)a	n	(%)a
Age in years, mean (SD)	75.4	(6.6)	74.9	(6.4)	76	(6.7)	75.8	(6.7)
Female	261	(46.2)	124	(47.0)	83	(43.7)	54	(48.6)
Race/ethnicity
Non-Hispanic White	454	(80.6)	223	(84.8)	151	(79.5)	80	(72.7)
Missing	2		1		0		1
At least some college education	370	(65.5)	163	(61.7)	135	(71.1)	72	(64.9)
Living alone	192	(34.2)	86	(32.7)	66	(34.9)	40	(36.4)
Missing	3		1		1		1
Difficulty in any IADL activity	129	(23.2)	55	(21.2)	48	(25.5)	26	(24.1)
Missing	10		5		2		3
Difficulties in 3+ ADL	38	(6.8)	15	(5.7)	14	(7.4)	9	(8.3)
Missing	5		3		0		2
Charlson comorbidity
0	342	(60.5)	157	(59.5)	117	(61.6)	68	(61.3)
1	79	(14.0)	34	(12.9)	29	(15.3)	16	(14.4)
2+	144	(25.5)	73	(27.7)	44	(23.2)	27	(24.3)
Duration of diabetes (years)
Mean (SD)	6	(6.1)	5.4	(5.6)	6	(6.5)	7.2	(6.6)
Median (IQR)	3.5	(1.3, 10.0)	3	(1.3, 8)	3.1	(1.3, 10.1)	5.6	(1.3, 11.9)
<1 y	106	(18.8)	50	(18.9)	35	(18.4)	21	(18.9)
1–<3 y	165	(29.2)	83	(31.4)	58	(30.5)	24	(21.6)
3–<5 y	55	(9.7)	29	(11.0)	18	(9.5)	8	(7.2)
5–<10 y	98	(17.3)	49	(18.6)	30	(15.8)	19	(17.1)
10+ y	141	(25.0)	53	(20.1)	49	(25.8)	39	(35.1)
Diabetes medications in 2 y prior to study entry
Insulin only	67	(11.9)	32	(12.1)	21	(11.1)	14	(12.6)
Metformin only	99	(17.5)	56	(21.2)	33	(17.4)	10	(9.0)
Sulfonylureas only	212	(37.5)	121	(45.8)	61	(32.1)	30	(27.0)
Mixed	176	(31.2)	47	(17.8)	73	(38.4)	56	(50.5)
Diabetes meds >2 y prior	11	(1.9)	8	(3)	2	(1.1)	1	(0.9)

Note: ADL = activities of daily living; IADL = instrumental activities of daily living; IQR = interquartile range.

^aPercentages were calculated among participants with nonmissing values.

Five hundred sixty-two individuals contributed 3 602 dementia-free years, and 132 individuals contributed 511 dementia follow-up years (mean 3.9 years after dementia diagnosis). There were 300 all-cause hospitalizations following dementia diagnoses, of which 93 were PPHs. During follow-up, 338 (60%) were hospitalized before dementia diagnosis or in the dementia-free group ([hereafter nondementia group] 122 were hospitalized once, 81 were hospitalized twice, and 135 were hospitalized for ≥3 times) vs 109 individuals (83%) after dementia diagnosis (30 were hospitalized once, 32 were hospitalized twice, and 47 were hospitalized ≥3 times). Twenty-three percent (n = 128) of the nondementia group had at least one PPH (83 had 1, 21 had 2, and 24 had ≥3 hospitalizations) compared with 42% (n = 55) after dementia diagnosis (32 had 1, 16 had 2, and 7 had ≥3 hospitalizations). PPHs due to cardiovascular events (cPPH) accounted for 46% in the nondementia group and 30% of all PPHs in the dementia group. On the other hand, PPHs due to diabetes (dPPH) and urinovolemic (uPPH) reasons accounted for 8% and 14% of all PPHs in the nondementia group and 17% and 28% in the dementia group, respectively. Other PPHs were similar between dementia and nondementia groups. Table 2 shows the differences in the total number of hospitalizations according to dementia status and HbA1c levels. The largest differences were observed in the number of overall PPH out of all-cause admissions and in the number of uPPH out of overall PPH admissions.

Table 2.

Total Number and Crude Rate (per 1 000 person-years) of Hospitalization Outcomes by HbA1c Levels Averaging Over 2-Year Periods and by Nondementia and Dementia Status During Study Follow-up

	HbA1c < 7%				HbA1c 7%–8%				HbA1c > 8%
	Nondementia (1 507 person-years)		Dementia (172 person-years)		Nondementia (1 364 person-years)		Dementia (192 person-years)		Nondementia (731 person-years)		Dementia (147 person-years)
Hospitalizations	N	Rate	N	Rate	N	Rate	N	Rate	N	Rate	N	Rate
All-cause	424	281.4	92	535.3	340	249.3	106	552.5	212	290.0	102	692.1
All PPHsa	92	61.1	24	139.6	66	48.4	34	177.2	52	71.1	35	237.5
Diabetes PPHb	6	4.0	3	17.5	6	4.4	8	41.7	5	6.8	5	33.9
Respiratory PPHc	24	15.9	7	40.7	17	12.5	5	26.1	14	19.2	9	61.1
Urinovolemic PPHd	9	6.0	4	23.3	9	6.6	11	57.3	11	15.1	11	74.6
CVD PPHe	46	30.5	10	58.2	28	20.5	11	57.3	22	30.1	7	47.5
Other acute PPHf	10	6.6	1	5.8	7	5.1	3	15.6	3	4.1	3	20.4

Notes: CVD = cardiovascular disease; PPH = potentially preventable hospitalization.

^aAll PPHs = diabetes, respiratory, urinovolemic, CVD, and other acute PPHs. ^bDiabetes PPH = diabetes-related hospitalizations including diabetes without mention of complication, diabetes with short-term or long-term complications, low-extremity amputation among people with diabetes, and hypoglycemia. ^cRespiratory PPH = respiratory-related hospitalizations including pneumonia, influenza, and asthma/chronic obstructive pulmonary disease (based on Agency for Healthcare Research and Quality definition). ^dUrinovolemic PPH = urinovolemic-related hospitalizations including urinary tract infection and dehydration. ^eCVD PPH = cardiovascular disease–related hospitalizations including congestive heart failure, hypertension, and angina. ^fOther acute PPH = other acute hospitalizations including appendix, cellulitis, gastric/duodenal/peptic ulcer, gastroenteritis, and seizure.

PPH Outcomes by Dementia Status

Figure 2 shows the crude and adjusted rates of hospitalizations according to dementia status. Among participants with dementia, the rate of overall PPH was 182 admissions per 1 000 person-years, more than 3 times the rate of those without dementia, which was 58 admissions per 1 000 person-years (crude RR: 3.12; 95% CI: 2.24–4.34). After adjustment for additional covariates, the RR was 2.27 (95% CI: 1.60–3.21). For PPHs due to diabetes and urinovolemic reasons, the crude admission rates were considerably higher among those with dementia (31 vs 5 per 1 000 person-years; crude RR: 6.63; 95% CI: 3.41–12.89 for dPPH, and 51 vs 8 per 1 000 person-years; crude RR: 6.32; 95% CI: 3.39–11.79 for uPPH). After full adjustment for covariates, the RR was 5.90 (95% CI: 2.70–12.88) for dPPH and 5.17 (95% CI: 2.49–10.73) for uPPH. The admission rate for respiratory reasons (rPPH) was also significantly higher among people with dementia, with a 2.01 adjusted RR (95% CI: 1.06–3.83). The admission rates for cPPH and oPPH were not significantly different by dementia status.

Figure 2.

Crude hospital admission rates per 1 000 person-years for all-cause and potentially preventable hospitalizations (PPHs) by dementia status and rate ratios (RRs) and 95% confidence intervals (CIs) estimates for the associations between incident dementia and rate of admissions. Total years of follow-up used in computation of rates were 3 602 y for the nondementia group and 511 for the dementia group. Fully adjusted models were adjusted for age, sex, race, education, living arrangement, deficits in instrumental activities of daily living, duration of diabetes, and Charlson comorbidity score. All PPHs included diabetes, respiratory, urinovolemic, cardiovascular disease (CVD), and other acute PPHs. Diabetes PPHs included diabetes without mention of complication, diabetes with short-term or long-term complications, low-extremity amputation among people with diabetes, and hypoglycemia. Respiratory PPHs included pneumonia, influenza, and asthma/chronic obstructive pulmonary disease (based on Agency for Healthcare Research and Quality definition). Urinovolemic PPHs included urinary tract infection and dehydration. CVD PPHs included congestive heart failure, hypertension, and angina. Other acute PPHs included appendix, cellulitis, gastric/duodenal/peptic ulcer, gastroenteritis, and seizure. PY = person-years; pre = nondementia group; post = dementia group.

Sensitivity Analyses

After excluding the 2-year follow-up period in which a person died, the association between dementia and admission rates was attenuated for uPPH (adjusted RR: 4.17; 95% CI: 1.87–9.29) and unchanged for dPPH (adjusted RR: 5.95; 95% CI: 2.67–13.27). When we further adjusted for a nursing home placement measure, the adjusted RRs were attenuated for overall PPH (RR: 1.51, 95% CI: 1.08–2.11), uPPH (RR: 3.11, 95% CI: 1.53–6.35), and dPPH (RR: 3.98, 95% CI: 1.76–9.01).

PPH Outcomes by Dementia Status and HbA1c Levels

Table 2 shows the crude rates of hospitalizations according to dementia status and HbA1c levels. Crude overall PPH admission rates were 61, 48, and 71 per 1 000 person-years in people with average HbA1c levels of <7%, 7%–8%, and >8% during nondementia follow-up, and 140, 177, and 238 per 1 000 person-years during postdementia follow-up, respectively. Similarly, crude uPPH admission rates were 6, 7, and 15 per 1 000 person-years in persons with average HbA1c levels of <7%, 7%–8%, and >8% during nondementia follow-up, and 23, 57, and 75 per 1 000 person-years during postdementia follow-up, respectively. Because we found no evidence that the association between HbA1c levels and PPH admission rates vary by dementia status (p values for interaction terms between dementia status and HbA1c levels were >.05 for all outcomes), the models were simplified to only include the main exposures for dementia status and HbA1c levels. Figure 3 shows the adjusted rates of hospitalizations according to HbA1c levels. The fully adjusted RR for overall PPH in people with >8% HbA1c was 1.43 (95% CI: 1.00–2.06), and for people with <7% HbA1c, it was 1.18 (95% CI: 0.85–1.65) compared with HbA1c of 7%–8%. The RR for uPPH was also different between >8% and <7% HbA1c level groups.

Figure 3.

Rate ratios (RRs) and 95% confidence intervals (CIs) estimates for the associations between HbA1c levels averaging over prior 2-y period and rate of admissions. Models were adjusted for age, sex, race, education, living arrangement, deficits in instrumental activities of daily living, duration of diabetes, Charlson comorbidity score, and dementia status. All potentially preventable hospitalizations (PPHs) included diabetes, respiratory, urinovolemic, cardiovascular disease (CVD), and other acute PPHs. Diabetes PPHs included diabetes without mention of complication, diabetes with short-term or long-term complications, low-extremity amputation among people with diabetes, and hypoglycemia. Respiratory PPHs included pneumonia, influenza, and asthma/chronic obstructive pulmonary disease (based on Agency for Healthcare Research and Quality definition). Urinovolemic PPHs included urinary tract infection and dehydration. CVD PPHs included congestive heart failure, hypertension, and angina. Other acute PPHs included appendix, cellulitis, gastric/duodenal/peptic ulcer, gastroenteritis, and seizure.

Sensitivity Analyses

We examined whether frailty altered the association between HbA1c level and hospitalization outcomes by including interaction terms between frailty and HbA1c levels in fully adjusted models. We found no evidence for such interactions (p > .05 for all).

Discussion

In this population-based prospective study of persons with diabetes, aged 65 and older, and who were initially dementia-free, we found significantly higher PPH admission rates among persons with dementia compared with those without dementia, especially for PPHs that involve diabetes complications, dehydration, and UTI. Overall in this population, 22% of hospitalizations before dementia and 31% after dementia were potentially preventable. We also found that, regardless of dementia status, poor glucose control of HbA1c >8% was associated with a borderline increase in rates of overall PPH relative to more moderate control of 7%–8% HbA1c, and increased rates of PPHs due to dehydration and UTI when compared with stricter control of <7% HbA1c levels. Importantly, these findings remained significant after adjusting for indicators for patient complexity, including comorbidity and frailty, and factors involved in diabetes management, such as a living arrangement and duration of diabetes. Since PPHs can signal opportunities for improvement in primary care management, our findings suggest that people with diabetes and dementia may benefit from an even more anticipatory model of care focusing on monitoring and avoiding complications that lead to hospitalizations.

Our observation that 31% of hospitalizations among patients with diabetes and dementia were PPHs is similar to the 28% comparable estimate reported in the study by Phelan et al. (16), which examined PPHs in a more historical cohort of ACT participants with dementia regardless of diabetes diagnosis and is somewhat higher than the 20% rate reported in the study by Kim et al. (20), that focused on older adults with diabetes but without dementia. In parallel with Phelan et al.’s study, patients with diabetes and dementia were at increased risk for PPHs due to UTI and dehydration, but unlike the general dementia population, patients with diabetes and dementia were also at increased risk for hospitalizations due to diabetes complications relative to their counterparts without dementia. The diabetes-related hospitalizations were mainly due to long-term diabetes complications, such as kidney, eye, neurological, circulatory, and other specified or nonspecified manifestations that accounted for about 70% of the admissions. Interestingly, in our study, dementia diagnosis was not associated with PPHs due to cardiovascular reasons. Another interesting observation is that when we excluded the follow-up period in which a person died, uPPH RR estimate by dementia status was attenuated by about 19%, while dPPH remained unchanged. This might be partially due to changes in the use of physician orders for life-sustaining treatment (POLST) in persons with dementia (21). As evidence accumulates that persons with POLST orders for comfort measures only (section B) are less likely to die in a hospital than persons with POLST order for a full treatment or without POLST form (22), our results may suggest that comfort might be more achievable in patients with diabetes complications than in patients with UTI and dehydration. This assertion, however, needs validation in future studies. Finally, our observation that higher nursing home placement rates among persons with dementia (36%) than those without (30%) accounted for a difference in several cause-specific preventable admissions is consistent with a recent report that showed that about 60% of community-dwelling patients with dementia are discharged to nonacute care facilities and these facilities are a major source of preventable admissions in this population (23). In fact, our analyses showed that nursing home placement was the strongest predictor of PPHs. As such, efforts directed at preventing hospitalizations in people with dementia might also delay their transition to a nursing care facility and prevent future admissions.

Our finding that poorer glycemic control of >8% HbA1c levels was potentially associated with heightened rates of overall and cause-specific PPHs is consistent with several previous studies (24,25). Our results also echo a recent ADA Standard of Care recommendations that suggest an HbA1c treatment goal of <8% in complex multimorbid patients with cognitive impairment (10). Although US Department of Veterans Affair (VA) and Centers for Medicare and Medicaid Services (CMS) claim and registry data suggest that patients with dementia are at risk for hypoglycemia (12), and tight HbA1c control of<7%, which was found in more than half of VA dementia patients, is associated with antidiabetic medications with a heightened risk of hypoglycemia (11), we found no evidence for a preponderance of tight glucose control and increased numbers of hypoglycemic events requiring hospitalization in our population. In fact, in our study, tight glucose control accounted for only one-third (34%) of dementia follow-up years, and only 6% of diabetes admissions were due to hypoglycemia. Our results instead suggested increased admission rates due to hyperglycemic complications such as dehydration and UTI in the poor glycemic control group relative to the tight control group. These conflicting results may be explained by our focus on hospital admissions rather than overall acute care utilization events, including emergency department visits that often capture less severe episodes of hypoglycemia.

There are several mechanisms by which dementia and poor glucose control can increase risks of PPHs in general and those that involve diabetes complications, dehydration, and UTI in particular. Ability to monitor glucose, adhere to the prescribed treatment regimen, and recognize hyperglycemic symptoms are almost certainly more difficult in the context of reduced cognition. In fact, people with advanced dementia may have difficulty independently managing their chronic conditions and must typically rely on family/friend caregivers. Caregivers may, in turn, face challenges in recognizing early symptoms of decompensation, which often involve atypical presentations, and taking swift actions for mitigating risks of further escalation (26). Another possible explanation is that people with dementia might actually be more vulnerable to metabolic abnormalities than their nondemented counterparts (27) because dementia and diabetes may act synergistically to increase the central nervous system vulnerability to hyperglycemia and lead to more severe neurocognitive manifestations of acute illness, in terms of early onset of delirium and functional impairment.

This work has 2 public health implications. Knowledge of the reasons for the most common PPHs may help primary care providers consider proper differential diagnosis and initiate early management protocols. This might help health care systems to reduce their currently escalating rates of preventable hospitalization among populations with dementia (23) and diabetes (28). Knowledge of the risk and benefits of different glycemic targets in this population might help clinicians optimize diabetes management and minimize risks of exposure to a potentially stressful and iatrogenic environment.

This study has several potential limitations. Given the preponderance of White participants in the study population, our results may not be broadly generalizable. Also, as an integrated health care delivery system, KPWA takes a proactive approach to manage and reduce a patient’s risk associated with various comorbid conditions. Therefore, the observed rates of hospitalization are likely to be lower than those in less integrated fee-for-service environments and could be considered as a lower bound of estimates. Next, while our glycemia exposure measure was based on thousands of measurements of HbA1c, these clinical laboratory measurements were obtained at irregular intervals. As patients come for visits at sporadic intervals, the irregular data pattern resemble usual clinical encounters. Furthermore, we did not access data for ambulatory care visits prior to admissions and, therefore, cannot conclude if outpatient care contacts occurred. Additionally, PPHs were defined by principle discharge diagnosis rather than defining preventability through chart reviews. Although adjudication of the preventability through chart review would be optimal to determine circumstances that led to the admission, this approach was beyond the scope of this project. Finally, it is likely that dementia diagnosis led to alterations in diabetes management toward either a stricter or more liberal level of control. As such, the results of this study, in terms of optimal glycemic targets, should serve as a basis for future clinical trials to confirm or refute the observational results. The strengths of this study include the prospective population-based design that systematically detects all incident dementia cases in a community-based cohort of older adults, access to clinical laboratory and pharmacy data, use of previously published models for glucose exposures (15), long follow-up before and after dementia diagnosis, and complete capture of hospitalizations through the EHR at KPWA.

In summary, our results that persons with comorbid dementia and diabetes have high rates of potentially avoidable hospitalizations suggest opportunities for improving care in this population through a more proactive and collaborative care models based on the provision of ongoing primary care (29). Proactive multidisciplinary care models should include enhanced medical surveillance through recurrent check-ins aimed at recognizing and, when indicated, treating incipient symptoms, and providing psychosocial support for caregivers of people with dementia. Our results also provide observational data to support the ADA recommendations of <8% HbA1c target in complex patients with coexisting chronic conditions.

Funding

This study was supported by funding from the National Institute on Aging (R03AG061305). The Adults Changes in Thoughts study is funded by a National Institutes on Aging Cooperative Grant (U01 AG006781).

Conflict of Interest

None declared.

Author Contributions

O.Z. and O.Y. designed the study, wrote the analysis plan, and interpreted the results. O.Y. performed analyses with feedback from O.Z., R.L.W., P.K.C., S.L.G., and E.B.L. O.Z. and O.Y. wrote the first draft of the manuscript with critical feedback and revisions from R.L.W., P.K.C., S.L.G., T.S., S.B., and E.B.L. O.Z., O.Y., R.L.W., P.K.C., S.L.G., T.S., S.B., and E.B.L gave final approval of the version to be published. O.Z. is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.