Abstract
Background and Aims:
Screening for substance use disorder (SUD) in general medical settings may be particularly important in patients with comorbid health conditions exacerbated by SUD. This study evaluated whether SUD is associated with type 2 diabetes mellitus (T2DM)-complications in patients with co-occurring T2DM and hypertension.
Design:
Analysis of a limited data set obtained through IBM Watson Health Explorys, a platform integrating data from electronic health records. Matched controls were defined for each of five SUDs: tobacco use disorder (TUD), opioid use disorder (OUD), cocaine use disorder, cannabis use disorder (CUD), and alcohol use disorder (AUD) using Mahalanobis distance within propensity score calipers.
Setting:
All patients were seen in the MetroHealth System (Cleveland, Ohio, USA) and had diagnosis codes for T2DM and hypertension.
Participants:
SUD group participants had a diagnosis of abuse/dependence for the substance of interest. Controls for each SUD group had no diagnosis code related to the SUD of interest and were selected to match the SUD patients on demographics, residential zip code median income, and body mass index. Total sample sizes for each SUD-control comparison ranged from 1,160 for CUD to 22,128 for TUD.
Measurements:
Outcome was diagnosis (yes/no) of four T2DM-complications (cerebrovascular accident, diabetic neuropathy, diabetic renal disease, myocardial infarction) and all-cause mortality.
Findings:
Logistic regressions revealed that SUD was significantly associated with greater risk of cerebrovascular accident (TUD-odds ratio (OR)=1.79, OUD-OR=1.94), cocaine use disorder-OR=2.67), diabetic neuropathy (TUD-adjusted odds ratio (aOR)=1.47, cocaine use disorder-aOR=1.35, AUD-aOR=1.27), diabetic renal disease (TUD-aOR=1.25, OUD-OR=1.34), myocardial infarction (TUD-OR=1.96, OUD-OR=2.01, cocaine use disorder-OR=2.68, CUD-OR=2.48, AUD-OR=1.42), and mortality (TUD-OR=1.15, cocaine use disorder-OR=1.61, CUD-aOR=1.49, AUD-OR=1.35).
Conclusions:
Among patients in Ohio USA with both type 2 diabetes mellitus (T2DM) and hypertension, those with substance use disorders appear to have greater risk for T2DM-complications and all-cause mortality.
Keywords: Type 2 Diabetes Mellitus, Hypertension, Substance use disorder (SUD), electronic health record (EHR)
INTRODUCTION
Substance use disorder (SUD) screening has been recommended for general medical healthcare settings but remains under-utilized (1). An alternative to universal SUD screening would be screening in patients with medical conditions significantly exacerbated by SUD. This value-based approach may be particularly useful for patients with multiple chronic health conditions, who account for the majority of healthcare spending and who interface frequently with the healthcare system, affording greater opportunity for screening and treatment coordination (2). However, understanding the relationship between SUD and medical complications is challenging since randomizing patients to SUD/control is neither ethical nor feasible and, thus, clinical research is limited to observational investigation. The increasing use of electronic health records (EHRs) provides large datasets for such investigation but have the disadvantage of potentially poor data quality and, as with all observational research, the potential for study confounds (3). Our research group has developed a method for evaluating the associations between SUD and medical complications, designed to reduce the potential for study confounds, by using an EHR data platform validated in past research and a combination of rigorous matching and analytic approaches to control for potential confounds. This paper presents the findings from our first application of this approach for which we selected patients with type 2 diabetes mellitus (T2DM) with co-occurring hypertension; this patient group was selected due to the prevalence, cost, and morbidity and mortality associated with this comorbidity (4). The association of T2DM-complications with SUD was evaluated for five SUDs; significant associations of all SUDS to all T2DM-complications would be consistent with poor treatment adherence generally associated with SUDs (5, 6) as a potential causal mechanism, while differences in associations across SUDs could indicate direct physiological effects of the abused substance.
METHODS
Setting
The study data were derived from the Explorys (IBM Watson Health; Cleveland, OH) technology platform, which utilizes a health data gateway server behind the firewall of participating healthcare organizations. The server collects data from a variety of health information systems (e.g., EHR, billing systems, lab/tests systems, etc.). The data are then de-identified and passed into the Explorys data grid, which is a private cloud-based datastore, and are standardized and normalized (7). The Explorys dataset has been validated in past research (7, 8). Our analytic approach, in which patients with SUD are “matched” to patients without SUD to decrease the possibility of study confounds, required a limited dataset available to us only from the MetroHealth System Explorys dataset. MetroHealth is an integrated healthcare system in Northeast Ohio that consists of 1 tertiary care academic medical center, 4 emergency departments, and over 2 dozen ambulatory clinics seeing approximately 25,000 inpatients per year with over 1.2 million outpatient visits annually. A data use agreement between MetroHealth and the University of Cincinnati (UC) was established to allow the use of limited datasets. This study was deemed not to be human subjects research by the UC institutional review board.
Study Population
Study participants were identified using ICD-9-CM and ICD-10-CM codes. The ICD-9/10 codes utilized for this study are provided in the supporting information (Tables S1–S3). The Explorys Cohort Discovery tool was used to identify adult MetroHealth patients with at least one ICD code of T2DM and of hypertension as an encounter diagnosis (i.e., see Table S1). At the time of Explorys query, the dataset contained approximately 17 years of patient data and 50,897 patients with T2DM and hypertension.
SUD groups.
SUD groups were identified for five SUDs: tobacco use disorder (TUD), opioid use disorder (OUD), cocaine use disorder, cannabis use disorder (CUD), and alcohol use disorder (AUD). The ultimate goal of this research is to identify patients whose medical outcomes can be improved by treating their SUD. To be included in an SUD group, a patient needed to have at least one encounter diagnosis code for abuse or dependence for the substance (see Table S2 for ICD-9/10 codes). Only abuse/dependence codes, as opposed to non-specific “use” codes, were used to identify SUD group participants; this follows the ICD coding recommended for SUDs outlined in the DSM-5 (9), and, thus, our SUD group participants would likely be candidates for SUD treatment. There were a minimal number of patients with a non-cocaine stimulant use disorder thus the analyses focused on patients with a cocaine abuse/dependence ICD-9/10 code.
Control groups.
For each of the five SUD groups, a control group was identified. To be included as a potential control participant, a patient could have neither an ICD-9/10 encounter diagnosis code of abuse/dependence nor an SUD-related code for the target substance; the SUD-related codes include acute intoxication, substance-induced disorders, and substance withdrawal (see Table S2 for ICD-9/10 code ranges). An SUD’s control group patients could have an SUD or SUD-related ICD-9/10 code for other substances – for example, OUD control participants could have encounter diagnosis codes for TUD, cocaine use disorder, CUD, and AUD; this allowed an evaluation of the specific SUD while controlling for other SUDs (see Data Analysis). Because patients with a given SUD may differ from patients without the given SUD on a variety of factors that could impact T2DM-complications, we created a control group that was matched to each SUD group. Matching variables included: age, sex, race, ethnicity, median income of zip code of residence, and body mass index (BMI).
The Ohio Valley Node (OVN) matching programs, which are SAS macros (version 9.4; Cary, NC) created by one of the authors (DL), were used for this study. The OVN matching programs and associated documentation are available at the National Drug Abuse Clinical Trials Network dissemination website (http://ctndisseminationlibrary.org/display/1287.htm); the programs are open-source and can be used free of charge. The OVN matching program uses Mahalanobis distance within propensity score calipers as the distance measure (10). Computational demand is reduced by applying propensity score calipers, using absolute propensity score difference, before matching to cull the set of potential matches. The matching process determines the similarity of an SUD and potential control participant by using the Mahalanobis calculation to coalesce the respective differences for each matching variable into a single distance measure. An optimal matching algorithm paired each SUD participant with a control participant so as to minimize the sum of the distances over the selected pairs (11).
T2DM-complication Variables
T2DM is associated with multiple, clinically significant, microvascular and macrovascular complications including: 1) cerebrovascular accident (CVA); 2) diabetic neuropathy; 3) diabetic renal disease; and 4) myocardial infarction (MI) (12). The ICD-9/10 codes used for each T2DM-complication category, along with the number of patients having each code, are provided in the supporting information (Table S3). While the data are correlational in nature (i.e., participants are not randomized to SUD/control groups), our approach helped to ensure that the T2DM-complication diagnoses occurred after the patient had the SUD of interest and, thus, that the SUD could have played a role in developing the complication. To this end, patients were scored as: 1) positive for the T2DM-complication if the initial T2DM-complication encounter diagnosis date was later than the initial dates for T2DM, hypertension, and SUD encounter diagnoses (controls were assessed using the initial SUD encounter diagnosis date of their matched SUD participant); 2) negative for the T2DM-complication if they did not have the T2DM-complication encounter diagnosis; or 3) missing if the initial T2DM-complication diagnosis date was not later than the initial dates for the T2DM/hypertension/SUD diagnoses. The rationale for using the missing code is based on the potential for patients to have a disorder prior to it being diagnosed and entered into the EHR. For example, a patient could have a cocaine use disorder for years before it is entered into the EHR. Hence, there is no way to ascertain whether a T2DM-complication diagnosis preceding a SUD diagnosis reflects the complication actually preceding the patient having the SUD or just preceding the entry of the SUD diagnosis into the EHR. The missing code reflects the unknown temporal relationship.
All-Cause Mortality
Social Security Death Index data is linked to Explorys patient information, and the Explorys system provides a patient’s year of death, if applicable. Mortality, unlike the T2DM complications described above, does not have the potential problems of temporal order (i.e., if a patient has died, we presume that all the encounter diagnoses of interest were made prior to the death). Patients with no indication of death were assumed to be alive at the time of data retrieval.
Data Analysis
All analyses were completed using SAS, Version 9.4. Statistical tests were conducted at an α level of 0.05 (two-tail) for all measures. For each SUD, logistic generalized mixed-model regressions (with a random effect to account for matched patient pairs) were used to evaluate whether SUD status was significantly associated with each of the five outcomes. Each regression potentially included other SUDs and demographic covariates (including time elapsed since initial T2DM diagnosis) selected using an automated procedure which performs a separate regression using each possible combination of potential nuisance covariates, and then selects the combination with the optimal resulting corrected Akaike Information Criterion (AIC-C); Tables S4–S8 list the covariates included in each regression. For the SUD nuisance covariates, participants with either abuse/dependence or SUD-related codes were scored as positive for the SUD. The proportion of data coded as missing (i.e., when initial T2DM-complication diagnosis date was not later than the initial dates for T2DM/hypertension/SUD diagnoses) ranged from a minimum of 5.1% (CVA – tobacco dataset) to 17.7% (diabetic neuropathy – opioid dataset). The median proportion of missing data was 7.4%. Crude odds ratio (OR) results were obtained from the comparison of the SUD group with its matched control without controlling for additional variables (e.g., other SUDs, length of time since diagnosis, etc.). Adjusted odds ratio (aORs) were yielded by logistic regressions comparing the SUD group with its matched control using AIC-C to determine inclusion of additional variables.
RESULTS
SUD prevalence was: 21.8% for TUD, 1.9% for OUD, 2.2% for cocaine use disorder, 1.1% for CUD, and 8.0% for AUD. As can be seen in Table 1, the matching procedure created comparable matched groups, with no significant differences. Table 2 displays the results of the comparisons between each SUD group and its matched control, including ORs and aORs. Discrepancies between ORs and aORs may reflect the more precise control provided by the logistic regression analysis and/or may reflect complicating factors such as the relatively low prevalence of a T2DM-complication/mortality and the number of variables selected by AIC-C in the statistical model. The most conservative interpretation of the findings is to consider a SUD-control group comparison to be significant based on the results of the logistic regression (i.e., as indicated by the aOR rows in Table 2) and to judge the degree to which the difference is clinically meaningful based on the smaller of the aOR/crude OR values (bolded in Table 2); the summary of findings in this section reflects this conservative approach.
Table 1.
Demographics for substance use disorder and matched controlled cohorts by substance use disorder
| Tobacco Use Disorder | Control (n=11064) | SUD (n=11064) | Test statistic (p-value)* |
|---|---|---|---|
| Age at time of data retrieval m(sd) | 61.0 (11.7) | 61.0 (11.6) | W = 0.3 (p = 0.79) |
| Female n (%) | 5845 (52.8%) | 5842 (52.8%) | χ2(1) = 0.0 (p = 0.97) |
| Race n (%) | χ2(2) = 0.0 (p = 1.00) | ||
| Black / African American | 5163 (46.7%) | 5165 (46.7%) | |
| White | 4592 (41.5%) | 4587 (41.5%) | |
| Other † | 1309 (11.8%) | 1312 (11.9%) | |
| Hispanic n (%) | 414 (3.7%) | 415 (3.8%) | χ2(1) = 0.0 (p = 0.97) |
| Median income (2006–2010) for the zip code of residence m(sd) | $37013.7 (14079.5) | $36981.4 (14246.9) | W = −0.5 (p = 0.62) |
| Average BMI m(sd) | 32.0 (9.6) | 31.9 (9.9) | T = 0.9 (p = 0.36) |
| Opioid Use Disorder | Control (n=965) | SUD (n=965) | Test statistic (p-value)* |
| Age at time of data retrieval m(sd) | 59.4 (10.2) | 59.4 (10.3) | W = 0.1 (p = 0.91) |
| Female n (%) | 440 (45.6%) | 440 (45.6%) | χ2(1) = 0.0 (p = 1.00) |
| Race n (%) | χ2(2) = 0.0 (p = 1.00) | ||
| Black / African American | 359 (37.2%) | 359 (37.2%) | |
| White | 480 (49.7%) | 479 (49.6%) | |
| Other † | 126 (13.1%) | 127 (13.2%) | |
| Hispanic n (%) | 48 (5.0%) | 48 (5.0%) | χ2(1) = 0.0 (p = 1.00) |
| Median income (2006–2010) for the zip code of residence m(sd) | $36596.1 (14009.9) | $36665.1 (14138.8) | W = −0.1 (p = 0.92) |
| Average BMI m(sd) | 31.3 (10.2) | 31.3 (10.5) | W = 0.0 (p = 0.98) |
| Cocaine Use Disorder | Control (n=1097)‡ | SUD (n=1097) | Test statistic (p-value)* |
| Age at time of data retrieval m(sd) | 57.7 (8.1) | 57.7 (8.0) | W = 0.2 (p = 0.87) |
| Female n (%) | 444 (40.5%) | 444 (40.5%) | χ2(1) = 0.0 (p = 1.00) |
| Race n (%) | χ2(2) = 0.0 (p = 1.00) | ||
| Black / African American | 728 (66.4%) | 729 (66.5%) | |
| White | 262 (23.9%) | 261 (23.8%) | |
| Other † | 107 (9.8%) | 107 (9.8%) | |
| Hispanic n (%) | 40 (3.6%) | 40 (3.6%) | χ2(1) = 0.0 (p = 1.00) |
| Median income (2006–2010) for the zip code of residence m(sd) | $33745.4 (12418.6) | $33739.9 (12525.7) | W = 0.2 (p = 0.87) |
| Average BMI m(sd) | 31.3 (10.0) | 31.2 (10.1) | W = 0.0 (p = 0.99) |
| Cannabis Use Disorder | Control (n=580) | SUD (n=580) | Test statistic (p-value)* |
| Age at time of data retrieval m(sd) | 54.0 (10.9) | 53.9 (10.9) | W = 0.2 (p = 0.85) |
| Female n (%) | 237 (40.9%) | 237 (40.9%) | χ2(1) = 0.0 (p = 1.00) |
| Race n (%) | χ2(2) = 0.0 (p = 1.00) | ||
| Black / African American | 325 (56.0%) | 325 (56.0%) | |
| White | 206 (35.5%) | 206 (35.5%) | |
| Other † | 49 (8.4%) | 49 (8.4%) | |
| Hispanic n (%) | 19 (3.3%) | 20 (3.4%) | χ2(1) = 0.0 (p = 0.87) |
| Median income (2006–2010) for the zip code of residence m(sd) | $34425.7 (12515.9) | $34400.3 (12658.1) | W = 0.2 (p = 0.87) |
| Average BMI m(sd) | 31.8 (9.9) | 31.9 (10.3) | W = 0.0 (p = 0.97) |
| Alcohol Use Disorder | Control (n=4060) | SUD (n=4060) | Test statistic (p-value)* |
| Age at time of data retrieval m(sd) | 62.0 (11.2) | 62.0 (11.3) | W = 0.1 (p = 0.95) |
| Female n (%) | 1335 (32.9%) | 1335 (32.9%) | χ2(1) = 0.0 (p = 1.00) |
| Race n (%) | χ2(2) = 0.0 (p = 1.0) | ||
| Black / African American | 1948 (48.0%) | 1948 (48.0%) | |
| White | 1648 (40.6%) | 1646 (40.5%) | |
| Other † | 464 (11.4%) | 464 (11.4%) | |
| Hispanic n (%) | 184 (4.5%) | 184 (4.5%) | χ2(1) = 0.0 (p = 1.00) |
| Median income (2006–2010) for the zip code of residence m(sd) | $35805.3 (13974.0) | $35803.1 (14118.2) | W = 0.0 (p = 1.00) |
| Average BMI m(sd) | 30.5 (9.5) | 30.5 (9.7) | W = 0.3 (p = 0.79) |
Quantitative measures were tested either with Wilcoxon rank-sum test (W) or Student’s t-test (T), depending on whether cohort variances tested significantly different; Categorical measures were tested with Pearson’s chi-square test of independence (χ2 (degrees of freedom)).
“Other” race includes: American Indian / Alaska Native, Asian, Multiracial, Native Hawaiian / Pacific Islander, and “missing” (i.e., no race data available);.
The Control cohort for the Cocaine SUD cohort excludes patients with cocaine use disorder and any other stimulant (e.g., amphetamine) use disorder.
Table 2.
Diabetes-related diagnosis prevalence in SUD vs. matched control cohorts and SUD odds ratios
| SUD* | Result | Cerebrovascular Accident | Diabetic Neuropathy | Diabetic Renal Disease | Myocardial Infarction | All-Cause Mortality |
|---|---|---|---|---|---|---|
|
Tobacco (N=22,128) |
Prevalence | 5.8% vs. 3.3% | 18.2% vs. 12.3% | 12.5% vs. 9.7% | 5.3% vs. 2.8% | 12.9% vs. 11.4% |
| Odds Ratio (95% CI) | 1.79 (1.56 – 2.04) | 1.60 (1.47 – 1.72) | 1.33 (1.22 – 1.45) | 1.96 (1.70 – 2.27) | 1.15 (1.06 – 1.24) | |
| aOR (95% CI) | 3.45 (2.70 – 4.40) | 1.47 (1.35 – 1.61) | 1.25 (1.14 – 1.38) | 5.79 (4.38 – 7.66) | 1.16 (1.06 – 1.27) | |
|
Opioid (N=1,930) |
Prevalence | 5.2% vs. 2.8% | 16.1% vs. 11.7% | 9.9% vs. 7.5% | 4.9% vs. 2.5% | 12.7% vs. 9.7% |
| Odds Ratio (95% CI) | 1.94 (1.18 – 3.18) | 1.44 (1.08 – 1.93) | 1.34 (0.96 – 1.88) | 2.01 (1.21 – 3.36) | 1.35 (1.02 – 1.8) | |
| aOR (95% CI) | 3.28 (1.23 – 8.74) | 1.34 (0.97 – 1.85) | 1.83 (1.04 – 3.21) | 6.10 (2.06 – 18.09) | 1.37 (0.99 – 1.91) | |
|
Cocaine (N=2,194) |
Prevalence | 7.7% vs. 3.0% | 19.7% vs. 12.3% | 12.4% vs. 9.4% | 5.2% vs. 2.0% | 12.0% vs. 7.8% |
| Odds Ratio (95% CI) | 2.67 (1.75–4.07) | 1.74 (1.35 – 2.24) | 1.36 (1.03 – 1.81) | 2.68 (1.60 – 4.48) | 1.61 (1.21 – 2.14) | |
| aOR (95% CI) | 10.91 (4.38 – 27.13) | 1.35 (1.00 – 1.83) | 1.27 (0.90 – 1.77) | 12.94 (3.57 – 46.94) | 1.95 (1.39 – 2.72) | |
|
Cannabis (N=1,160) |
Prevalence | 4.3% vs. 2.6% | 16.7% vs. 10.7% | 10.9% vs. 9.0% | 4.8% vs. 2.0% | 8.3% vs. 5.7% |
| Odds Ratio (95% CI) | 1.69 (0.86 – 3.33) | 1.68 (1.16 – 2.44) | 1.24 (0.83 – 1.86) | 2.48 (1.21 – 5.09) | 1.49 (0.94 – 2.36) | |
| aOR (95% CI) | 1.71 (0.42 – 7.02) | 1.35 (0.86 – 2.12) | 1.39 (0.86 – 2.23) | 14.70 (2.74 – 78.86) | 2.49 (1.18 – 5.23) | |
|
Alcohol (N=8,120) |
Prevalence | 5.7% vs. 3.9% | 18.1% vs. 12.8% | 12.1% vs. 10.3% | 4.9% vs. 3.5% | 17.0% vs. 13.1% |
| Odds Ratio (95% CI) | 1.47 (1.18 – 1.81) | 1.50 (1.32 – 1.71) | 1.20 (1.04 – 139) | 1.42 (1.13 – 1.78) | 1.35 (1.19 – 1.53) | |
| aOR (95% CI) | 1.35 (0.92 – 1.98) | 1.27 (1.10 – 1.46) | 1.07 (0.91 – 1.27) | 1.62 (1.06 – 2.47) | 1.54 (1.32 – 1.78) |
SUD=Substance Use Disorder; First row for each SUD has prevalence of the diagnosis in the SUD versus matched control cohorts. Second row for each SUD has the unadjusted OR with 95% confidence intervals; a shaded cell indicates a statistically significant (p < 0.05) effect. Third row for each SUD has adjusted OR (aOR) with 95% confidence intervals from logistic regression; a shaded cell indicates a statistically significant (p < 0.05) effect. Bolded value is the smaller value of the OR/aOR; the N= total sample for the analysis (SUD + matched controls).
The analyses for TUD revealed significantly worse outcomes for the TUD group for CVA (5.8% vs. 3.3%; OR=1.79 (1.56–2.04)), diabetic neuropathy (18.2% vs. 12.3%; aOR=1.47 (1.35–1.61)), diabetic renal disease (12.5% vs. 9.7%; aOR=1.25 (1.14–1.38)), MI (5.3% vs. 2.8%; OR=1.96 (1.70–2.27)), and mortality (12.9% vs. 11.4%; OR=1.15 (1.06–1.24)). For OUD, there were significantly worse outcomes for the OUD group for CVA (5.2% vs. 2.8%; OR=1.94 (1.18–3.18)), diabetic renal disease (9.9% vs. 7.5%; OR=1.34 (0.96–1.88), and MI (4.9% vs. 2.5%; OR=2.01 (1.21–3.36)). The cocaine analyses revealed significantly worse outcomes for the cocaine use disorder group for CVA (7.7% vs. 3.0%; OR=2.67 (1.75–4.07)), diabetic neuropathy (19.7% vs. 12.3%; aOR=1.35 (1.00–1.83)), MI (5.2% vs. 2.0%; OR=2.68 (1.60–4.48)), and mortality (12.0% vs. 7.8%; OR=1.61 (1.21–2.14)). The CUD analyses revealed significantly worse outcomes for the CUD group for MI (4.8% vs. 2.0%; OR=2.48 (1.21–5.09)) and mortality (8.3% vs. 5.7%; aOR=1.49 (0.94–2.36)). Finally, the AUD analyses revealed significantly worse outcomes for the AUD group for diabetic neuropathy (18.1% vs. 12.8%; aOR=1.27 (1.10–1.46)), MI (4.9% vs. 3.5%; OR=1.42 (1.13–1.78)), and mortality (17.0% vs. 13.1%; OR=1.35 (1.19–1.53)).
DISCUSSION
This paper reports findings from an EHR data analysis evaluating the association between SUDs and T2DM-complications in patients with T2DM and hypertension for five SUDs (TUD, OUD, cocaine use disorder, CUD, and AUD) using a rigorous procedure to define matched controls. While research has evaluated the association between tobacco use, and to a lesser degree alcohol use, and T2DM-complications, this is, to our knowledge, the first published report of the association for patients with T2DM and hypertension and the first to evaluate the association between T2DM-complications and other specific SUDs. The results suggest that in patients with T2DM and hypertension, those with an SUD may be at significantly greater risk for several T2DM-complications and death.
Our findings revealed significantly higher rates of CVA, diabetic neuropathy, diabetic renal disease, MI, and mortality in patients with TUD, which is consistent with the well-established adverse effects of tobacco (12, 13). Comparing the effect sizes from the current study to those of past research is challenging in that studies have varied in the T2DM-complications evaluated and how they were defined. In addition, to our knowledge, this is the first published study focusing on a T2DM patient sample with comorbid hypertension. The most comparable data come from a meta-analysis conducted by Pan and colleagues, which found a pooled relative risk (RR) for tobacco use of 1.54 (1.41 – 1.69) for stroke, 1.51 (1.42–1.62), for coronary heart disease, and 1.55 (1.46–1.64) for total mortality (14). Converting the ORs from the present study to RRs, our findings indicate a RR of 1.75 for CVA, 1.91 for MI, and 1.13 for mortality, which are generally consistent with the RRs reported by Pan and colleagues.
The present findings revealed that patients with AUD were more likely than control patients to have experienced diabetic neuropathy, MI, and mortality. The increased prevalence of diabetic neuropathy is consistent with some prior research suggesting that heavy alcohol consumption increases risk of developing neuropathy in people with diabetes (15). However, chronic alcohol use is also a risk factor for peripheral neuropathy, independent of diabetic neuropathy (16); the present analyses included only T2DM-specifc diagnoses (see Table S3), but misclassification of this outcome may have occurred in the clinical decision to attribute neuropathy to T2DM or another etiology. The findings of increased risk for MI and mortality are consistent with the results of a study by Blomster and colleagues (17) finding a significantly increased, dose-dependent, risk for cardiovascular events and all-cause mortality among T2DM patients reporting heavy drinking, relative to patients reporting no drinking. The present results are inconsistent with those from an analysis of data from Medicare and Medicaid beneficiaries conducted by Leung and colleagues (18) which revealed that those with AUD, relative to those without AUD, did not differ significantly on the incidence of neuropathy. In addition, unlike the present study, which did not find significant differences between patients with and without AUD on cerebrovascular disease and diabetic renal disease, Leung et al. (18) found that individuals with AUD, relative to patients without AUD, were more likely to have cerebrovascular disease and less likely to have nephropathy. The differences in results may reflect differences in: the patient populations studied, with the present study including patients with hypertension and patients beyond Medicare and Medicaid beneficiaries; the procedures used to identify the control group, with the present study using a rigorous matching procedure; or other study differences.
To our knowledge, the present study is the first to explore the association between T2DM-complications and OUD, CUD, and cocaine use disorder. The present findings suggest an increased risk for diabetic renal disease, CVA, and MI in patients with T2DM and hypertension who have an OUD. Though the association between OUD and diabetic renal disease in this analysis was relatively small, the results are consistent with postulated mechanisms by which chronic opioid use may cause kidney injury and damage (19). Prescription opioid use has been hypothesized as a risk factor for both adverse cerebrovascular and cardiovascular effects, but a recent analysis of a large prospective cohort study found no association between prescription opioid use and stroke, but an association was detected for death attributed to cardiovascular disease (20). Our results are also consistent with past research finding that the long-term use of opioid therapy increases the risk of MI (21). Despite the current opioid overdose epidemic, we did not detect a significant difference in mortality between OUD patients and matched controls.
In the present study, patients with diagnoses of T2DM and hypertension who also had a CUD diagnosis were at significantly greater risk for MI and death. These findings are consistent with prior research finding that cannabis use increases the risk for MI (22) and death attributed to hypertension (23). The present findings suggest that patients with T2DM and hypertension who also have a cocaine use disorder are at increased risk for CVA and MI, which is consistent with the physiological effects of cocaine and greater prevalence of these events observed in cocaine-using populations (24, 25). The present findings also suggest that patients with cocaine use disorder had a greater mortality risk than their matched controls. This finding represents a contribution to the literature in that a meta-analysis concluded that the literature on mortality among cocaine users is limited, and suggested that researchers use routinely collected data to help answer research questions about cocaine and mortality (26). The present findings also suggest that there may be an increased risk of diabetic neuropathy in patients with T2DM and hypertension who have cocaine use disorder. While there have been a few case reports relating cocaine use to neuropathies (27–29) our study is, to our knowledge, the first to the demonstrate an association between cocaine use disorder and diabetic neuropathy.
While the present results suggest that patients with T2DM and hypertension who also have an SUD may be at significantly greater risk for several T2DM-complications and death relative to those without an SUD, the findings need to be replicated before being given serious clinical consideration, given that many of the findings are the first of their kind in the published literature. If replicated, these findings would indicate the importance of screening for SUD in patients with T2DM/hypertension and providing and/or referring patients to appropriate SUD treatment (see Walter et al. (6) for a discussion of SUD screening and treatment in patients with T2DM). The present results, which did not find significant associations for all SUDs with all T2DM-complications, may indicate that biological mechanisms, as opposed to simple non-adherence to treatment, serve to increase the risk of adverse medical outcomes; if these results are replicated, an exploration of potential physiological mechanisms may be of interest.
This study has several strengths. First, as noted by Leung et al. (18), T2DM-complications tend to develop over a longer period of time and, thus, EHR analyses evaluating T2DM-complications should include multi-year data but many EHRs have been established relatively recently, and, thus cannot provide this degree of longitudinal data. The inclusion of data from multiple years is a definite study strength - the average number of years post-T2DM-diagnosis ranged from a low of 7.2 years for the CUD comparison to a high of 7.5 for the OUD comparison. The large sample sizes, ranging from a low of 1,160 for the CUD comparisons to a high of 22,128 for TUD is another study strength, as are the use of rigorous matching and analytic approaches. This study also has important limitations. First, the findings are correlational in nature and, thus, cause and effect determinations cannot be made. Second, while the sample sizes were relatively large, the sample was limited to patients being treated in the MetroHealth System, located in Northeast Ohio, and the extent to which the results are generalizable to the rest of the U.S. and other countries is unknown. Third, there may be other potential confounding factors that were not accounted for in the models. For example, the analyses could not control for potentially important factors, such as physical exercise, due to their lack of ready availability in the EHR. Fourth, our definition of control participants was based on substance abuse/dependence and SUD-related ICD-9/ICD-10 codes and did not include other potential indicators of problematic substance use (e.g., alcoholic hepatitis/cirrhosis diagnosis, etc.); hence, some control participants may have actually had problematic substance use, which would serve to reduce the differences observed between the SUD and control groups. Finally, while the Explorys dataset used is of relatively high quality, there is still the potential for data quality issues. For example, as is the case with medical conditions generally, the under-diagnosis, misclassification, or under-coding of SUD by clinicians is possible (30). Indeed, the proportions of patients with various SUDs in the current study are lower than the proportions reported in a recent SUD screening study conducted in primary care (31). Hence, the present findings might not generalize to patients identified in systematic SUD screening efforts.
In conclusion, the results suggest that in patients with T2DM and hypertension, those with an SUD may be at significantly greater risk for several T2DM-complications and all-cause mortality. Future research to replicate this finding and to delineate the potential mechanisms by which SUD may impact T2DM-related outcomes seems warranted.
Supplementary Material
Funding Source:
National Institute on Drug Abuse (NIDA) Clinical Trials Network: the Ohio Valley Node Network (Grant UG1DA013732)
Footnotes
Declarations of interest: TW, JT, and DL declare no conflicts of interest. DCK is the Chief Medical Informatics Officer of the MetroHealth System. In exchange for contributing de-identified data to the Explorys network, the MetroHealth System receives access to the Explorys Cohort Discovery tool, which was used to conduct this study. Neither DCK nor the MetroHealth System have any direct financial ties to Explorys (IBM Watson Health).
REFERENCES
- 1.Pace CA, Uebelacker LA. Addressing Unhealthy Substance Use in Primary Care. The Medical clinics of North America. 2018;102(4):567–86. [DOI] [PubMed] [Google Scholar]
- 2.Wu LT, Brady KT, Spratt SE, Dunham AA, Heidenfelder B, Batch BC, et al. Using electronic health record data for substance use Screening, Brief Intervention, and Referral to Treatment among adults with type 2 diabetes: Design of a National Drug Abuse Treatment Clinical Trials Network study. Contemporary clinical trials. 2016;46:30–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Groeneveld PW, Rumsfeld JS. Can Big Data Fulfill Its Promise? Circulation Cardiovascular quality and outcomes. 2016;9(6):679–82. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Wang G, Zhou X, Zhuo X, Zhang P. Annual Total Medical Expenditures Associated with Hypertension by Diabetes Status in U.S. Adults. American Journal of Preventive Medicine. 2017;53(6):S182–S9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Sansone RA, Sansone LA. Alcohol/Substance misuse and treatment nonadherence: fatal attraction. Psychiatry. 2008;5(9):43–6. [PMC free article] [PubMed] [Google Scholar]
- 6.Walter KN, Wagner JA, Cengiz E, Tamborlane WV, Petry NM. Substance Use Disorders among Patients with Type 2 Diabetes: a Dangerous but Understudied Combination. Current diabetes reports. 2017;17(1):2. [DOI] [PubMed] [Google Scholar]
- 7.Kaelber DC, Foster W, Gilder J, Love TE, Jain AK. Patient characteristics associated with venous thromboembolic events: a cohort study using pooled electronic health record data. Journal of the American Medical Informatics Association : JAMIA. 2012;19(6):965–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Pfefferle KJ, Gil KM, Fening SD, Dilisio MF. Validation study of a pooled electronic healthcare database: the effect of obesity on the revision rate of total knee arthroplasty. European journal of orthopaedic surgery & traumatology : orthopedie traumatologie. 2014;24(8):1625–8. [DOI] [PubMed] [Google Scholar]
- 9.American Psychiatric Association. Diagnostic and statistical manual of mental disorders: DSM-5. Fifth ed. Arlington VA: American Psychiatric Association; 2013. [Google Scholar]
- 10.Hintze J Data Matching – Optimal and Greedy. Kaysville, UT: 2007. [cited 2016 May]; Available from: http://www.webcitation.org/75BECFvI7. [Google Scholar]
- 11.Stuart EA. Matching methods for causal inference: A review and a look forward. Statistical science : a review journal of the Institute of Mathematical Statistics. 2010;25(1):1–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Cade WT. Diabetes-related microvascular and macrovascular diseases in the physical therapy setting. Physical therapy. 2008;88(11):1322–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Centers for Disease Control and Prevention . National Diabetes Statistics Report, 2017. Atlanta, GA: Centers for Disease Control and Prevention, US Department of Health and Human Services, 2017. [Google Scholar]
- 14.Pan A, Wang Y, Talaei M, Hu FB. Relation of Smoking With Total Mortality and Cardiovascular Events Among Patients With Diabetes Mellitus: A Meta-Analysis and Systematic Review. Circulation. 2015;132(19):1795–804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Emanuele NV, Swade TF, Emanuele MA. Consequences of alcohol use in diabetics. Alcohol Health & Research World. 1998;22(3):211–9. [PMC free article] [PubMed] [Google Scholar]
- 16.Zeng L, Alongkronrusmee D, van Rijn RM. An integrated perspective on diabetic, alcoholic, and drug-induced neuropathy, etiology, and treatment in the US. Journal of pain research. 2017;10:219–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Blomster JI, Zoungas S, Chalmers J, Li Q, Chow CK, Woodward M, et al. The Relationship Between Alcohol Consumption and Vascular Complications and Mortality in Individuals With Type 2 Diabetes. Diabetes care. 2014;37(5):1353–9. [DOI] [PubMed] [Google Scholar]
- 18.Leung G, Zhang J, Lin W-C, Clark RE. Behavioral Disorders and Diabetes-Related Outcomes Among Massachusetts Medicare and Medicaid Beneficiaries. Psychiatric Services. 2011;62(6):659–65. [DOI] [PubMed] [Google Scholar]
- 19.Mallappallil M, Sabu J, Friedman E, Salifu M. What Do We Know about Opioids and the Kidney? International Journal of Molecular Sciences. 2017;18(1):223. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Khodneva Y, Muntner P, Kertesz S, Kissela B, Safford MM. Prescription Opioid Use and Risk of Coronary Heart Disease, Stroke, and Cardiovascular Death Among Adults from a Prospective Cohort (REGARDS Study). Pain Medicine. 2016;17(3):444–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chou R, Turner JA, Devine EB, Hansen RN, Sullivan SD, Blazina I, et al. The effectiveness and risks of long-term opioid therapy for chronic pain: a systematic review for a National Institutes of Health Pathways to Prevention Workshop. Annals of internal medicine. 2015;162(4):276–86. [DOI] [PubMed] [Google Scholar]
- 22.Mittleman MA, Lewis RA, Maclure M, Sherwood JB, Muller JE. Triggering myocardial infarction by marijuana. Circulation. 2001;103(23):2805–9. [DOI] [PubMed] [Google Scholar]
- 23.Yankey BA, Rothenberg R, Strasser S, Ramsey-White K, Okosun IS. Effect of marijuana use on cardiovascular and cerebrovascular mortality: A study using the National Health and Nutrition Examination Survey linked mortality file. European journal of preventive cardiology. 2017;24(17):1833–40. [DOI] [PubMed] [Google Scholar]
- 24.Gorelick DA. Cocaine use disorder in adults: Epidemiology, pharmacology, clinical manifestations, medical consequences, and diagnosis. In: Post TW, editor. UpToDate. Waltham, MA: UpToDate; 2017. [Google Scholar]
- 25.Mittleman MA, Mintzer D, Maclure M, Tofler GH, Sherwood JB, Muller JE. Triggering of Myocardial Infarction by Cocaine. Circulation. 1999;99(21):2737–41. [DOI] [PubMed] [Google Scholar]
- 26.Degenhardt L, Singleton J, Calabria B, McLaren J, Kerr T, Mehta S, et al. Mortality among cocaine users: A systematic review of cohort studies. Drug and alcohol dependence. 2011;113(2):88–95. [DOI] [PubMed] [Google Scholar]
- 27.de Souza A, Desai PK, de Souza RJ. Acute multifocal neuropathy following cocaine inhalation. Journal of Clinical Neuroscience. 2017;36:134–6. [DOI] [PubMed] [Google Scholar]
- 28.Lee BY, Keane DJ. Atypical presentation of weakness: cocaine-induced peripheral neuropathy. Muscle & Nerve. 2011;44(4):643–4. [Google Scholar]
- 29.Stuard WL, Gallerson BK, Robertson DM. Alterations in corneal nerves following crack cocaine use mimic diabetes-induced nerve damage. Endocrinology Diabetes and Metabolism Case Reports. 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wu LT, Ghitza UE, Batch BC, Pencina MJ, Rojas LF, Goldstein BA, et al. Substance use and mental diagnoses among adults with and without type 2 diabetes: Results from electronic health records data. Drug and alcohol dependence. 2015;156:162–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Wu LT, McNeely J, Subramaniam GA, Brady KT, Sharma G, VanVeldhuisen P, et al. DSM-5 substance use disorders among adult primary care patients: Results from a multisite study. Drug and alcohol dependence. 2017;179:42–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.