Intensification of Diabetes Medications at Hospital Discharge and Clinical Outcomes in Older Adults in the Veterans Administration Health System

Key Points Question What is the association between intensification of outpatient diabetes medications at hospital discharge and clinical outcomes in older adults hospitalized for common medical conditions? Findings In this cohort study of 5296 propensity-matched older veterans with diabetes who were hospitalized for common medical conditions, discharge with intensified diabetes medication was associated with an increased risk of severe hypoglycemia within 30 days and was not associated with a reduction in severe hyperglycemia events or hemoglobin A1c level at 1 year. Meaning These findings indicate that short-term hospitalization may not be an effective time to intervene in long-term diabetes management.


Introduction
Modification of older adults' home medications during short-term hospitalization is common.
Changes to home medications may be temporary in response to acute illness or may reflect planned changes to management of chronic disease. [1][2][3][4][5] A particularly common scenario is adjustment of diabetes medications. 5,6 During hospitalization for acute illness, older adults with diabetes may experience fluctuating blood glucose control, driven by changes in eating patterns, medication exposures, and catecholamine surges. As a result, hospitalization is a high-risk time for serious hypoglycemia and hyperglycemia events. [7][8][9][10][11] Practice guidelines advise broader ranges for inpatient blood glucose levels than are recommended in the outpatient setting and recommend stopping the use of home oral agents and initiating the use of short-term insulin in many clinical scenarios. [12][13][14] Safe diabetes management for older adults requires careful balancing of the short-term risks of medication-induced hypoglycemia [15][16][17] with the long-term benefits of blood glucose control. 18,19 Transient elevations in blood glucose during hospitalization likely have little long-term significance, yet these increases commonly precipitate intensification of home regimens, including discharging patients with new insulin or oral agents. 5 For patients with uncontrolled diabetes, hospitalization could be an opportune time to address hyperglycemia and set patients on the path toward improved chronic disease control. However, the posthospitalization period is also a particularly high-risk time for adverse drug events and medication errors. Although clinical trials and guidelines have helped inform long-term diabetes treatment strategies for ambulatory older adults, this evidence does not reflect the perihospitalization period, during which time older adults have an increased susceptibility to adverse drug events 20,21 and may not be generalizable to hospitalized older adult populations, who face greater frailty and more limited life expectancy than clinical trial participants. 22,23 Because the clinical outcomes associated with diabetes medication intensifications made at hospital discharge are unknown, we conducted a retrospective cohort study of older adults with diabetes who were hospitalized in the national Veterans Health Administration (VHA) health system for common noncardiac conditions. We evaluated the association between intensification of home diabetes medications at hospital discharge and postdischarge outcomes, including severe hypoglycemia and hyperglycemia events, mortality, hemoglobin A 1c (HbA 1c ) control at 1 year, and persistent use of discharge medications at 1 year after discharge. discharged to the community setting (eFigure 1 in the Supplement). Diabetes was defined by the presence of 2 outpatient diagnoses or any hospital discharge diagnosis of diabetes in the 2 years that preceded the index hospitalization using previously validated algorithms. 24,25 Because diagnosisbased algorithms may capture patients with a history of diabetes or currently being evaluated for diabetes, to enhance specificity we examined only patients who were taking a diabetes medication before hospitalization or had an HbA 1c level greater than 6.5% (to convert to proportion of total hemoglobin, multiply by 0.01) in the year before hospitalization.
Conditions were identified by primary discharge diagnosis code grouped by Clinical Classification Software categories and included the following: acute coronary syndrome, arrhythmia, asthma, chest pain, chronic obstructive pulmonary disease, coronary artery disease, conduction disorders, heart failure, heart valve disorders, pneumonia, sepsis, skin infection, stroke, transient ischemic attack, and urinary tract infection. These conditions were chosen because they are among the most common medical discharge diagnoses for older adults and their short-term management does not typically require intensification of outpatient diabetes medications. Patients discharged with a secondary discharge diagnosis of diabetic ketoacidosis or nonketotic hyperglycemichyperosmolar coma were excluded because these conditions typically necessitate an immediate change in diabetes treatment.
To ensure accurate classification of medication use, we excluded patients likely to receive medications outside the VHA, including patients who received more than 20% of their outpatient care outside the VHA, patients admitted from skilled nursing facilities, and patients who had been hospitalized in the 30 days that preceded the index hospitalization. 26 Patients enrolled in hospice were excluded given differing goals of care. Because instructions to modify insulin dosing are infrequently accompanied by a new prescription, dosing changes cannot be accurately assessed using pharmacy databases; thus, we limited our study to patients not using insulin before hospitalization.

Exposure
We compared patients discharged with intensified diabetes medication regimens to those discharged without intensifications. Intensifications were defined as newly prescribed diabetes medications that were not being used before hospitalization and medications present on admission for which a discharge prescription was filled for a dose increase of more than 20%. Intensifications were ascertained based on VHA pharmacy dispensing data using previously published methods, which included medications filled within 2 days before to 2 days after discharge. 26,27 We examined all medication classes in use during the study period: biguanides, sulfonylureas, thiazolidinediones, α-glucosidase inhibitors, dipeptidyl peptidase 4 inhibitors, meglitinides, glucagon-like peptide 1 (GLP-1) agonists, sodium-glucose cotransporter 2 (SGLT2) inhibitors, and insulins.

Outcomes
The 2 primary outcomes were chosen a priori to assess possible benefits and harms of diabetes medication intensification: severe hyperglycemia events and severe hypoglycemia events. Primary outcomes were examined at 30 days to assess immediate outcomes and at 365 days to assess longer term outcomes. On the basis of prior studies, 15,28,29 primary outcomes were defined as a composite of emergency department (ED) visits, observation stays, and hospitalizations for severe hypoglycemia and severe hyperglycemia (eTable 1 in the Supplement). Secondary outcomes included all-cause readmissions at 30 and 365 days, mortality at 30 and 365 days, change in HbA 1c at 1 year, and persistent use of diabetes medication prescriptions filled at discharge at 1 year.
to clinical expertise 18 and included demographic characteristics, comorbidities, 31 prehospitalization and hospital vital signs, laboratory values, health care use, and medications (eTable 2 in the Supplement). Missing data were imputed using the fully conditional specification method and 20 imputation sets. One-to-one nearest neighbor matching without replacement was performed and covariate balance between groups was assessed using standardized mean differences. 32,33 Second, within propensity score-matched groups, survival analyses were conducted for hyperglycemia, hypoglycemia, mortality, and readmission outcomes using Cox proportional hazards regression models for mortality and Fine and Gray proportional subdistribution hazards models for all other outcomes to account for the competing risk of death. 34 For all models, SEs accounted for clustering of patients within hospitals. To aid in interpretation of subdistribution hazard models, unadjusted event rates are presented for each group.
Third, within propensity score-matched groups, the change in HbA 1c at 1 year after discharge was estimated using a difference-in-differences approach. Linear regression models were used to estimate the change in HbA 1c level associated with discharge with intensified diabetes medications after subtracting the background change among patients who did not receive medication intensifications. 35,36 Fourth, for the unmatched cohort, we examined persistence to diabetes medications filled at discharge during the subsequent year. We examined diabetes medication prescriptions filled at discharge, including new medication prescriptions and prescription fills of admission medications at higher, lower, or the same doses. For each diabetes medication filled at discharge, we calculated persistence as the number of days between the discharge fill and the last refill for the same or greater dose plus the days supplied by the latest refill. 37 We constructed Kaplan-Meier curves and used the log-rank test to examine differences in persistence by type of fill: continuation, dose increase, dose decrease, new oral medication, or new insulin.
We conducted subgroup analyses to determine the differential impact of exposure to intensified diabetes medications by prehospitalization diabetes control. We classified patients as having controlled or elevated prehospitalization HbA 1c levels using a threshold HbA 1c of 7.5%, acknowledging the uncertainty surrounding exact HbA 1c targets in older adults. 18,22,38 We then repeated propensity score matching and analyses for each baseline HbA 1c group separately.
Analyses were conducted using Stata software, version 14.1 (StataCorp LLC). For all analyses, we determined statistical significance using 95% CIs. A 2-sided P < .05 was also considered statistically significant.

Results
The unmatched cohort included 28 198 older adults with diabetes admitted to 115 VHA hospitals patients with a similar propensity score who did not receive intensifications (95.7% match rate).
Matched groups were well balanced on propensity score distribution and baseline characteristics (standardized mean differences for all covariates, <0.1) ( Table 1; eTable 3 and eFigure 2 in the Supplement).   a Selected covariates are presented; a full list of covariates included in the propensity score is given in eTable 2 in the Supplement.
c Balance between the groups was assessed before and after matching by comparing SMDs for each variable for which a difference of less than 0.10 was considered to indicate adequate balance.

Persistent Use of Diabetes Medications
In   Absolute change in HbA 1c , %

Received discharge intensification
Analysis includes the 4215 patients in the propensity-matched cohort who met the inclusion criteria for the HbA 1c analysis (outpatient HbA 1c measured between 6 and 18 months after the index hospitalization discharge). Postbaseline HbA 1c level was assessed before censoring as the HbA 1c level recorded at the closest date to 1 year after index hospitalization discharge, within the range of 6 to 18 months after discharge. Absolute change in HbA 1c calculated as postdischarge HbA 1c level minus preadmission HbA 1c level.
Complete difference-in-differences results are given in eTable 4 in the Supplement. The horizontal lines in the center of each box indicate the median; the lower and upper bounds of each box, the 25th and 75th percentiles; the lower and upper error bars, the most extreme value between the 25th percentile minus 1.5 times the interquartile range and the most extreme value between the 75th percentile plus 1.5 times the interquartile range.

Prehospitalization Baseline Hemoglobin A 1c Subgroup Analyses
Propensity score matching yielded a cohort of 2672 patients with controlled preadmission HbA 1c levels (Յ7.5%) and a cohort of 2524 patients with elevated preadmission HbA 1c levels (>7.5%), each equally split between those who received intensifications and those who did not. Covariate balance between groups in each cohort was excellent except for differences in the regional distribution of patients (eTables 5 and 6 and eFigure 3 in the Supplement).
Among matched patients with controlled baseline HbA 1c levels, the mean (SD) prehospitalization HbA 1c level was 6.8% (0.5%) for the intensified and not intensified groups. No differences were found in severe hypoglycemia events, severe hyperglycemia events, or secondary clinical outcomes among patients with controlled baseline HbA 1c levels who were discharged with or without diabetes medication intensifications ( Table 3; eTable 7 in the Supplement).
Among matched patients with elevated baseline HbA 1c levels, the mean (SD) prehospitalization HbA 1c level was 9.1% (1.5%) for the intensified group and 9.1% (1.6%) for the not intensified group.
No differences were found in severe hypoglycemia or hyperglycemia events among patients with elevated baseline HbA 1c levels who were discharged with or without diabetes medication intensifications (  Persistence analysis includes 18 455 patients who filled 1 or more diabetes medication prescriptions at discharge. A total of 24 085 unique medication prescriptions were included because patients could fill multiple diabetes medication prescriptions at discharge. For each unique medication prescribed at discharge, we calculated persistence as the number of days between the discharge prescription and the last refill for the same or greater dose plus the days supplied by the latest refill. If a patient died during the follow-up period, persistence was truncated at date of death. We reported persistence to 12 months after discharge by type of treatment, and to avoid undercounting because of transient nonadherence, we assessed refill history for 18 months.

Discussion
In this cohort study of older adults with diabetes who were hospitalized for common medical conditions, intensification of diabetes medications at hospital discharge was associated with increased short-term risk of severe hypoglycemia events without reduction in risk of severe hyperglycemia events or improvement in HbA 1c control at 1 year. Moreover, nearly half of discharge intensifications were not continued at 1 year. Despite the lack of association with improved diabetes control, older adults receiving diabetes medication intensifications at discharge had a lower risk of mortality at 30 days but no difference in mortality at 1 year. These results suggest intensification of older adults' outpatient diabetes medications during unrelated hospitalizations should generally be avoided. confound the association between discharge intensification and discharge outcomes. 6 Our study builds on these prior studies 6,39 by examining postdischarge outcomes in the VHA, a large, national, integrated health system, which allows for the inclusion of richer clinical characteristics and complete identification of postdischarge events that occur inside and outside the VHA. In addition, we examined a more recent period than prior studies 6,39 ; thus, differences in findings may in part reflect differences in diabetes medication classes.
In a secondary analysis, we observed that older adults receiving diabetes intensifications had a substantially lower risk of death in the first 30 days of discharge. This unexpected finding was consistent in elevated but not controlled HbA 1c subgroups and merits further examination in future studies. This finding contrasts with the Canadian study, 6 which found that discharge with new insulin was associated with an increased risk of death, and a prior trial 40 of intensive inpatient blood glucose control among critically ill patients, which found more intensive blood glucose control was associated with higher 90-day mortality. We anticipate this finding may be attributable to unmeasured suggests, readmission for hyperglycemia is rare, occurring in less than 0.5% of cohort patients with elevated baseline HbA 1c levels. Adverse drug events are typically highest in the initial weeks of treatment, and this risk is likely to be multiplied in the postdischarge period, during which older adults are typically exposed to multiple medication changes and hospital-associated disability. 20,21,42 Additional research is needed to determine best management practices for older adults with diabetes whose primary reason for hospitalization requires short-term treatment with medications known to greatly increase blood glucose levels, which will be continued to the outpatient setting (eg, corticosteroids for respiratory or autoimmune disease flairs) because these patients may benefit from short-term monitored intensifications.

Limitations
Our study has several limitations. First, it took place in the VHA health care system, a national integrated system that serves a predominately male population with greater multimorbidity that may not be generalizable to the entire US. Second, because we focused on older adults who are at higher risk of adverse drug events, our findings are not generalizable to younger populations. Third, although our study was strengthened by examining both VHA and Medicare data, we were not able to identify hyperglycemia and hypoglycemia events for which patients did not seek emergency care; thus, these events are likely underestimated. 43 Fourth, because of limitations of pharmacy data, we were unable to examine the impact of changes in insulin dosing and thus did not examine patients taking insulin before hospitalization. Pharmacy claims do not allow for the identification of discontinued medications; thus, medication classes started as substitutions for other classes were included in the study as intensifications. Fifth, as an observational study, there is a risk of unmeasured confounding by variables not included in the propensity score-matched analyses. Sixth, subgroup analyses were exploratory and may have been underpowered to demonstrate differential associations of diabetes medication intensification across baseline HbA 1c levels.

Conclusions
Among older adults hospitalized for common medical conditions, discharge with intensified diabetes medications was not associated with reduced severe hyperglycemia events or HbA 1c levels within 1 year but was associated with an increased risk of severe hypoglycemia events within 30 days. For