Abstract
Aims:
To identify substance use disorder (SUD) patterns and their association with T2DM health outcomes among patients with type 2 diabetes and hypertension.
Methods:
We used latent class analysis on electronic health records from the MetroHealth System (Cleveland, Ohio) to obtain the target SUD groups: i) only tobacco (TUD), ii) tobacco and alcohol (TAUD), and iii) tobacco, alcohol, and at least one more substance (PSUD). A matching program with Mahalanobis distance within propensity score calipers created the matched control groups: no SUD (NSUD) for TUD and TUD for the other two SUD groups. The numbers of participants for the target-control groups were 8,009 (TUD), 1,672 (TAUD), and 642 (PSUD).
Results:
TUD was significantly associated with T2DM complications. Compared to TUD, the TAUD group showed a significantly higher likelihood for all-cause mortality (adjusted odds ratio (aOR) = 1.46) but not for any of the T2DM complications. Compared to TUD, the PSUD group experienced a significantly higher risk for cerebrovascular accident (CVA) (aOR = 2.19), diabetic neuropathy (aOR = 1.76), myocardial infarction (MI) (aOR = 1.76), and all-cause mortality (aOR = 1.66).
Conclusions:
The findings of increased risk associated with PSUDs may provide insights for better management of patients with T2DM and hypertension co-occurrence.
Keywords: Electronic health records (EHR), Type 2 diabetes mellitus (T2DM), Hypertension, Polysubstance use disorder (PSUD), Health outcomes, All-Cause mortality
1. Introduction
Screening for substance use disorder (SUD) in general medical settings is important for health conditions that may be exacerbated by SUD. Such screening may be particularly important in 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 [1]. A recent study evaluating the association between individual SUDs and type 2 diabetes mellitus (T2DM)-complications in patients with co-occurring T2DM and hypertension found that individuals with a SUD had greater risk for T2DM-complications and all-cause mortality [2]. While of interest, the findings do not address the potential additional risk arising from polysubstance use. Relative to research on individual SUDs, there is a dearth of research on polysubstance use although there is evidence to suggest that approximately 11% of individuals with an SUD have a co-occurring alcohol or SUD and that polysubstance use is associated with an increased risk of several adverse outcomes [3]. Some research has evaluated the most common patterns of polysubstance use [4, 5] and suggests that alcohol, cannabis and/or tobacco, and alcohol, tobacco, and cannabis with or without other substances are among the common patterns. To our knowledge, an analysis of common polysubstance use patterns has not been conducted in patients with co-occurring T2DM and hypertension. The present study had two goals. The first was to utilize latent class analysis to delineate SUDs that commonly co-occur in a sample of patients with co-occurring T2DM and hypertension. The second goal was to evaluate the association of the resulting polysubstance use disorder patterns and T2DM-complications.
2. Methods
2. 1. Data
The study data were extracted from the cloud-computing platform Explorys, originally developed by Cleveland Clinic and then acquired by IBM Watson Health. The platform uses a health data gateway server, which connects to multiple data sources of participating healthcare organizations, to collect health data from different health information systems such as electronic health records (EHR), billing systems, lab/tests systems, etc. The collected data are de-identified before passing to the Explorys data grid, and then the data are standardized and normalized [6]. The Explorys dataset has been validated in past research [6, 7]. The present study methods required a limited dataset, which was available from the MetroHealth System. Under the integrated healthcare system in Northeast Ohio, MetroHealth has one tertiary-level academic medical center, four emergency departments, and more than two dozen ambulatory clinics, and has about 25,000 hospitalizations and 1.2 million outpatient visits yearly. The use of limited datasets has been facilitated by an agreement between MetroHealth and the University of Cincinnati (UC). This study was deemed not to be human subjects research by the UC institutional review board.
2.2. Study population
Study participants were identified using the same ICD-9 and ICD-10 codes used by Winhusen and colleagues [2]. The adult MetroHealth patients with at least one diagnosis of T2DM and hypertension were identified by the Explorys Cohort Discovery tool. At the time of Explorys query, the dataset contained approximately 17 years of patient data and 50,897 patients with T2DM and hypertension diagnoses.
2.3. Latent class analysis (LCA)
Latent class analysis (LCA) is a statistical approach that identifies subgroups/classes of sample subjects based on a set of indicator variables [8, 9]. These subgroups are mutually exclusive, and subjects within the subgroups share common characteristics. We used LCA in our sample of patients with type 2 diabetes and hypertension to identify possible subgroups based on their status on five SUD indicators: 1) tobacco use disorder (TUD); 2) alcohol use disorder (AUD); 3) opioid use disorder (OUD),4) cocaine use disorder; and 5) cannabis use disorder (CUD). A maximum of five LCA models corresponding to one to five subgroups can be fitted for five indicator variables. One subgroup would include the entire study sample and is not relevant given the goal of identifying SUD patterns associated with adverse outcomes. Thus, we evaluated four models, which included two to five subgroups. The best model was identified by comparing the models using the low Akaike Information Criteria (AIC) and the Bayesian Information Criteria (BIC) [8, 10]. LCA analysis was carried out in the R system [11].
2.4. SUD groups
The patterns of SUDs obtained from the LCA were used to heuristically define the observable SUD groups as only tobacco use disorder (TUD), tobacco and alcohol use disorder (TAUD), and tobacco, alcohol, and at least one more substance (PSUD). The second and third groups represented polysubstance co-occurrence. To evaluate the T2DM health risks, the TUD target (n=8,009) group was compared to the matched controls of no substance use disorder (NSUD, n=8,009), while TAUD (n=1,672) and PSUD (n=642) were compared to the matched patients with TUD.
2.5. Control groups
For each of the three SUD groups, a control group was selected from a pool of potential control patients. The first inclusion criterion for a potential control patient was that the patient could have neither an ICD-9/10 encounter diagnosis code of abuse/dependence nor a SUD-related code for the target substances using the same ICD-9 and ICD-10 codes used by Winhusen and colleagues [2]. For the TUD group, the control group included only those patients who did not have any SUD (NSUD). For the TAUD and PSUD groups, control patients included the patients with TUD. For each SUD group, a control group was selected by matching each patient in the SUD group to a single control patient using an optimal matching algorithm from the Ohio Valley Node matching programs (written in SAS version 9.4, Carry, NC) [2]. Age, sex, race, ethnicity, the median income of zip code of residence, and body mass index (BMI) were used as matching variables. Before performing the matching algorithm, propensity scores were calculated using the matching variables, and calipers were used to screen out pairs of patients with dissimilar propensity scores. The matching algorithm measured the similarity between patients within each prospective pair using a Mahalanobis distance calculated from the within-pair differences for the respective matching variables. For each of the three SUD cohorts, an optimal set of matched pairs was obtained by matching each SUD patient to a single, unique control patient so as to minimize the sum of the Mahalanobis distances over the selected pairs [12].
2.6. T2DM health outcomes
There are several clinically significant microvascular and macrovascular health outcomes, including i) cerebrovascular accident (CVA); ii) diabetic neuropathy; iii) diabetic renal disease; and iv) myocardial infarction (MI), associated with T2DM [13]. The status of T2DM health outcomes for each of the adult MetroHealth patients was extracted from the EHR by ensuring that the T2DM health outcome was diagnosed after the patient had the SUD diagnosis of interest (controls were assessed by the SUD encounter date of their matched SUD participants). 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 diagnosis of interest and, thus, that the SUD could have played a role in developing the complication. The ICD-9/10 codes used for each T2DM-complication category are provided in the supporting information (Table S1). For a detailed description of the extraction procedures, please refer to the study by Winhusen and colleagues [2].
2.7. All-cause mortality
Social Security Death Index data are linked to Explorys and 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 death). Patients with no indication of death were assumed to be alive at the time of data retrieval.
2.8. Data analysis
All analyses except LCA were completed using SAS, Version 9.4. Statistical tests were conducted at an α level of 0.05 (two-tail) for all measures. After matching, the test of equality of the characteristics measured in continuous scale, including age, median income, and BMI, over the targeted SUDs versus respective controls were conducted by the normal approximation of the Wilcoxon rank-sum test. The categorical characteristics, including sex, race, ethnicity, were tested by using the chi-square (χ2) test.
For each of the three SUD groups, and each of the five outcomes of interest, a logistic regression was performed with the outcome as the response variable and an SUD-group-vs.-control-group indicator as the covariate of interest. Each regression also included, as nuisance covariates, the set of demographic variables for adjustment. Table 3 contains the adjusted odds ratio (aOR) for the covariate of interest for each regression.
3. Results
3.1. Latent class analysis
SUD diagnoses prevalence was: 21.78% for TUD, 1.9% for OUD, 2.2% for cocaine use disorder, 1.1% for CUD, and 8.0% for AUD. The latent class analysis identified three subgroups as being optimal based on the lowest values of AIC and BIC (Table 1). A review of the three subgroups from the 3-class model (Figure 1) were interpreted as: 1) No use or low prevalence of TUD diagnoses (which was 74% of the sample); 2) TUD and low prevalence of AUD diagnoses (which was 23.6% of the sample), and TUD, AUD, at least one more SUD diagnoses (which was 2.4% of the sample).
Table 1.
Model selection criteria for 2-5 latent subgroups from 5 SUDs
| Model Selection Criteria | Number of Subgroups |
|||
|---|---|---|---|---|
| 2 | 3 | 4 | 5 | |
| AIC | 101059.4 | 100865.9 | 100879.9 | 100878.6 |
| BIC | 101156.6 | 101016.1 | 101083.2 | 101134.9 |
Figure 1.
Latent patterns of SUDs among patients with type 2 diabetes mellitus and hypertension
3.2. Polysubstance use disorder patterns and T2DM-complications
The prevalence of the three observable SUD groups was: 15.7% for TUD, 3.3% for TAUD, and 1.3% for PSUD. The test results to compare the sociodemographic characteristics of the SUD groups with the matched control groups are summarized in Table 2. No significant difference was evident in any of the characteristics between the groups, which supports their comparability.
Table 2.
Demographics for substance use disorder and matched controlled cohorts by substance use disorder
| Tobacco Use Disorder Only (TUD) | Control (n=8,009) | TUD (n=8,009) | Test statistic (p-value)* |
|---|---|---|---|
| Age at time of data retrieval m(sd) | 61.6 (12.1) | 61.6 (12.1) | Z = 0.23 (p = 0.82) |
| Female n (%) | 4,665 (58.2%) | 4,660 (58.2%) | χ2(1) = 0.01 (p = 0.94) |
| Race n (%) | χ2(2) = 0.01 (p = 1.00) | ||
| Black / African American | 3,576 (44.6%) | 3,575 (44.6%) | |
| White | 3,408 (42.5%) | 3,405 (42.5%) | |
| Other † | 1,025 (12.8%) | 1,025 (12.8%) | |
| Hispanic n (%) | 306 (3.8%) | 307 (3.8%) | χ2(1) = 0.0 (p = 0.97) |
| Median income (2006-2010) for the zip code of residence m(sd) | $37,690 (14,428) | $37,686 (14,613) | Z = 0.33 (p = 0.74) |
| Average BMI m(sd) | 33.6 (9.8) | 32.6 (10.1) | Z = 0.37 (p = 0.71) |
| Tobacco & Alcohol Use Disorder (TAUD) | TUD (n=1,672) | TAUD ((n=1,672) | Test statistic (p-value)* |
|
| |||
| Age at time of data retrieval m(sd) | 61.42 (10.0) | 61.52 (10.3) | Z = −0.40 (p = 0.69) |
| Female n (%) | 584 (34.9%) | 582 (34.8%) | χ2(1) = 0.01 (p = 0.94) |
| Race n (%) | χ2(2) = 0.01 (p = 1.00) | ||
| Black / African American | 817 (48.9%) | 819 (49.0%) | |
| White | 693 (41.4%) | 691 (41.3%) | |
| Other † | 162 (9.7%) | 162 (9.7%) | |
| Hispanic n (%) | 63 (3.8%) | 63 (3.8%) | χ2(1) = 0.00 (p = 1.00) |
| Median income (2006-2010) for the zip code of residence m(sd) | $35,382 (12,753) | $35,550 (13,297) | Z = −0.35 (p = 0.72) |
| Average BMI m(sd) | 29.67 (8.8) | 29.53 (8.9) | Z = 0.43 (p = 0.67) |
| Polysubstance use disorders (PSUDs) | TUD (n=642) | PSUDs (n=642) | Test statistic (p-value)* |
|
| |||
| Age at time of data retrieval m(sd) | 57.88 (8.1) | 57.84 (8.3) | Z = 0.06 (p = 0.95) |
| Female n (%) | 234 (36.4%) | 233 (36.3%) | χ2(1) = 0.00 (p = 0.95) |
| Race n (%) | χ2(2) = 0.06 (p = 0.97) | ||
| Black / African American | 364 (56.7%) | 364 (56.7%) | |
| White | 238 (37.1%) | 236 36.8%) | |
| Other † | 40 (6.2%) | 42 (6.5%) | |
| Hispanic n (%) | 12 (1.9%) | 14 (2.2%) | χ2(1) = 0.16 (p = 0.69) |
| Median income (2006-2010) for the zip code of residence m(sd) | $33,942 (12,327) | $34,036 (12,790) | Z = 0.07 (p = 0.95) |
| Average BMI m(sd) | 29.59 (8.9) | 29.43 (8.9) | Z = 0.36 (p = 0.72) |
Tests were performed by normal approximation with Wilcoxon rank-sum test. For all other variables, chi-square (χ2) test was applied.
“Others” race includes American Indian/Alaska Native, Asian, Multiracial, Native Hawaiian/Pacific Islander, and “missing” (i.e., no race data available).
The prevalence of the microvascular and macrovascular health outcome diagnoses (T2DM complications) and all-cause mortality and adjusted odds ratios (aORs) of the SUDs are presented in Table 3. Higher prevalence along with significantly higher odds of CVA (PR = 7.4% vs. 4.9%; aOR = 1.55), diabetic neuropathy (PR = 19.7% vs. 14.2%; aOR = 1.48), diabetic renal disease (PR = 13.5% vs. 11.8%; aOR = 1.16), MI (PR = 7.5% vs. 3.8%; aOR = 2.11), and all-cause mortality (PR = 12.51% vs. 11.34%; aOR = 1.14) were evident for the TUD group compared to the matched control NSUD group. In comparison to the TUD group, the TAUD group did not show any significant difference in risks for the microvascular and macrovascular health outcome diagnoses. However, the TAUD group showed a 46% higher likelihood of all-cause mortality (PR = 15.8% vs. 11.7%; aOR = 1.46) compared to the patients with only TUD.
Table 3.
Diabetes-related diagnosis prevalence in SUD vs. matched control cohorts and SUD odds ratios
| SUD vs. Control* | Result† | Cerebrovascular Accident | Diabetic Neuropathy | Diabetic Renal Disease | Myocardial Infarction | All-Cause Mortality |
|---|---|---|---|---|---|---|
| TUD vs. Control | Prevalence | 7.4% vs. 4.9% | 19.7% vs. 14.2% | 13.5% vs. 11.8% | 7.5% vs. 3.8% | 12.5% vs. 11.3% |
| aOR (95% CI) | 1.55 (1.35-1.77) | 1.48 (1.35-1.61) | 1.16 (1.05-1.28) | 2.11 (1.82-2.44) | 1.14 (1.03-1.27) | |
|
| ||||||
|
TAUD vs. TUD
|
Prevalence | 8.2% vs. 8.1% | 22.2% vs. 21.2% | 14.6% vs. 14.4% | 7.0% vs. 8.0% | 15.8% vs. 11.7% |
| aOR (95% CI) | 0.98 (0.76-1.27) | 1.04 (0.87-1.24) | 1.02 (0.84-1.25) | 0.85 (0.65-1.11) | 1.46 (1.18-1.80) | |
|
| ||||||
|
PSUD vs. TUD
|
Prevalence | 11.2% vs. 5.5% | 28.0% vs. 18.6% | 16.1% vs. 14.7% | 11.2% vs. 6.6% | 12.8% vs. 8.3% |
| aOR (95% CI) | 2.19 (1.43-3.37) | 1.76 (1.33-2.32) | 1.13 (0.82-1.54) | 1.76 (1.16-2.66) | 1.66 (1.14-2.41) | |
SUD = Substance use disorder, Control = No substance use disorder, TUD = tobacco use disorder only, TAUD = tobacco and alcohol use disorder, PSUD = Polysubstance (tobacco, alcohol, and at least one of opioid, cocaine or cannabis) use disorder.
Prevalence = Prevalence of the diagnosis in the SUD versus matched control cohorts, aOR = adjusted OR with 95% confidence intervals from logistic regression; a shaded cell indicates a statistically significant (p < 0.05) effect.
Significantly worse outcomes were evident for the patients with PSUD (tobacco, alcohol, and at least one more SUD of either OUD, cocaine use disorder, or CUD). On top of the risk of TUD, patients with PSUD experienced significantly higher risk for CVA (PR = 11.2% vs. 5.5%; aOR = 2.19), diabetic neuropathy (PR = 28.0% vs. 18.6%; aOR = 1.76), MI (PR = 11.2% vs. 6.6%; aOR = 1.76), and all-cause mortality (PR = 12.8% vs. 8.3%; aOR = 1.66). However, risks for diabetic renal disease was not significantly different between the PSUD and TUD groups.
4. Discussion
This study evaluated the risk of T2DM microvascular and macrovascular complications, including CVA, diabetic neuropathy, diabetic renal disease and MI, and all-cause mortality for patients with T2DM and hypertension and SUDs, including TUD, TAUD, and PSUD. These SUD groups were defined by applying latent class analysis (LCA). LCA has been used by several studies to study patterns of substance abuse [5, 14, 15] and the associated characteristics of the defined subgroups. In our study, a three-subgroup latent model was identified as the best model to identify SUD patterns. These subgroups can be labeled as: 1) no use/very low prevalence of tobacco; 2) tobacco and low prevalence of alcohol; and 3) tobacco, alcohol, and other substance(s). Past studies have examined polysubstance use patterns in diverse populations [16]. Although the patterns may vary across the study populations, geography, demographics, study structure, study time, types and number of considered substances [3], our LCA subgroup formation was consistent with most of the previous literature [5]. The present study is unique in contributing the patterns of SUDs among patients with T2DM and hypertension to the literature. Furthermore, this study used these latent subgroups [17] to form observed groups to study the association with T2DM health outcome diagnoses and all-cause mortality. This delineation of SUD patterns will facilitate the identification of vulnerable groups by healthcare providers and, thus, allows for potential intervention to improve outcomes in these patient populations.
The effect of individual SUDs on the same health outcomes in patients with T2DM/hypertension in the same setting was first demonstrated by Winhusen et al. [2]. Based on the LCA identified groups, this study most importantly tried to reveal how the risk of poor health outcomes and mortality is added for polysubstance users (i.e., our TAUD and PSUD groups) compared to TUD alone. As the negative health outcomes of tobacco are well documented in the literature [18–21], our study concentrated on comparing the TAUD group and PSUD group to a TUD control group. To our knowledge, this is the first study to evaluate the association between polysubstance use and T2DM health outcome diagnoses and all-cause mortality and, thus, the finding of heightened risk for adverse outcomes should be replicated before being considered clinically actionable. The observed low prevalence of SUD groups in this study, especially the PSUD (TAUD: 3.3%; PSUD: 1.3%), suggest that, if the findings are replicated, potential intervention efforts should be feasible, given the relatively modest number of patients, which serves to limit provider burden.
As a preliminary finding, this study showed the significantly higher risk of TUD on the T2DM health outcome diagnoses, including CVA, diabetic neuropathy, diabetic renal disease and MI, and all-cause mortality, compared to matched NSUD controls, which is consistent with the findings of the first study of this kind by Winhusen et al. [2]. The effect of TAUD compared to TUD on CVA, diabetic neuropathy, diabetic renal disease, and MI diagnoses were not found to be statistically significant, which indicated that the joint effect of alcohol and tobacco might incur the same risk as resulted from tobacco alone. Somewhat surprisingly, the prevalence of neuropathy was not statistically higher in the TAUD relative to the TUD group, which might be due to our inclusion of the diabetic neuropathy but not the alcoholic neuropathy ICD codes for this analysis. The diabetic neuropathy finding is consistent with the findings of Leung et al. [22], who did not find a statistically significant difference in diabetic neuropathy between alcohol users and non-users. The present study found a greater risk of all-cause mortality in the TAUD, relative to TUD, groups. A recent study by Hongli et al. [23] also revealed the joint effect of smoking and alcohol consumption on increasing risk of all-cause mortality. As the authors used non-smokers/non-drinkers as the reference, they did not show how alcohol added the risk on top of the risk of tobacco.
The present study also found evidence of a significantly higher risk for CVA, diabetic neuropathy, MI, and all-cause mortality of the T2DM and hypertension patients with PSUD compared to those with TUD. A comparison cohort is not available in the literature; however, as individual SUDs, the association of opioid use disorder, cocaine use disorder, cannabis use disorder with T2DM complications, and death was demonstrated by Winhusen et al. [2].
The study has several strengths. First, T2DM-complications tend to develop over a longer period of time and, thus, EHR analyses evaluating T2DM-complications should include multi-year data (this study had up to 17 years worth of 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. Second, the patterns of SUDs were identified by the LCA, a mixture model to identify the unobserved latent groups, and postulated these to close observable groups. Third, two-step adjustment of covariates, including matching program and logistic regression, were applied to evaluate the association attributed only to the SUD groups of interest. Fourth, to our knowledge, this is the first study to demonstrate the effect of PSUD relative to TUD on T2DM microvascular and macrovascular health outcome diagnoses and all-cause mortality.
This study also has some limitations. First, the patients were chosen from the MetroHealth System, located in Northeast Ohio, and thus the findings may not generalize to the other parts of the U.S. or other countries. Second, there may be other potential factors (e.g., physical exercise, treatment adherence) associated with health outcomes that were not included in the models and thus, not controlled for. Finally, although the data quality of Explorys is relatively high, there is a possibility of under-diagnosis, misclassification, or under-coding of SUDs [24].
In conclusion, the results suggest that in patients with T2DM and hypertension, the most common patterns of SUDS are tobacco use disorder (TUD) only, TUD with alcohol use disorder (TAUD), and TAUD with another co-occurring SUD (PSUD). Relative to TUD, patients with TAUD have an increased risk of all-cause mortality while patients with PSUD have an increased risk of several T2DM health complications and all-cause mortality.
Supplementary Material
Funding:
This study was funded by the National Institute on Drug Abuse (NIDA) Clinical Trials Network: the Ohio Valley Node Network (Grant UG1DA013732).
Footnotes
Conflict of interest: MTFK, TJW, 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).
Data availability:
The data underlying this article will be shared on reasonable request to the corresponding author.
References
- 1.Wu LT, 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. Contemp Clin Trials 46 (2016) 30–38, 10.1016/j.cct.2015.11.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Winhusen T., et al. , Medical complications associated with substance use disorders in patients with type 2 diabetes and hypertension: electronic health record findings. Addiction 114(8) (2019) 1462–1470, 10.1111/add.14607 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Crummy EA, et al. , One Is Not Enough: Understanding and Modeling Polysubstance Use. Front Neurosci 14 (2020) 569, 10.3389/fnins.2020.00569 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Connor JP, et al. , Polysubstance use: diagnostic challenges, patterns of use and health. Curr Opin Psychiatry 27(4) (2014) 269–75, 10.1097/YCO.0000000000000069 [DOI] [PubMed] [Google Scholar]
- 5.Halladay J, et al. , Patterns of substance use among adolescents: a systematic review. Drug and alcohol dependence (2020) 108222, 10.1016/j.drugalcdep.2020.108222 [DOI] [PubMed] [Google Scholar]
- 6.Kaelber DC, et al. , Patient characteristics associated with venous thromboembolic events: a cohort study using pooled electronic health record data. J Am Med Inform Assoc 19(6) (2012) 965–72, 10.1136/amiajnl-2011-000782 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Pfefferle KJ, et al. , Validation study of a pooled electronic healthcare database: the effect of obesity on the revision rate of total knee arthroplasty. Eur J Orthop Surg Traumatol 24(8) (2014) 1625–8, 10.1007/s00590-014-1423-2 [DOI] [PubMed] [Google Scholar]
- 8.Collins LM and Lanza ST, Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. Vol. 718. 2009: John Wiley & Sons. [Google Scholar]
- 9.Hagenaars JA and McCutcheon AL, Applied latent class analysis. 2002, Cambridge ; New York: Cambridge University Press. xxii, 454 p. [Google Scholar]
- 10.Nylund KL, Asparouhov T, and Muthén BO, Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural equation modeling: A multidisciplinary Journal 14(4) (2007) 535–569, 10.1080/10705510701575396 [DOI] [Google Scholar]
- 11.R Core Team, R: A Language and Environment for Statistical Computing. 2020, R Foundation for Statistical Computing: Vienna, Austria. [Google Scholar]
- 12.Stuart EA, Matching methods for causal inference: A review and a look forward. Stat Sci 25(1) (2010) 1–21, 10.1214/09-STS313 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Cade WT, Diabetes-related microvascular and macrovascular diseases in the physical therapy setting. Phys Ther 88(11) (2008) 1322–35, 10.2522/ptj.20080008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Lee H., et al. , Substance use patterns among adolescents: a latent class analysis. Journal of the American Psychiatric Nurses Association 26(6) (2020) 586–594, 10.1177/1078390319858658 [DOI] [PubMed] [Google Scholar]
- 15.Evans BE, Kim Y, and Hagquist C, A latent class analysis of changes in adolescent substance use between 1988 and 2011 in Sweden: associations with sex and psychosomatic problems. Addiction 115(10) (2020) 1932–1941, 10.1111/add.15040 [DOI] [PubMed] [Google Scholar]
- 16.Lorvick J, et al. , Polydrug use patterns, risk behavior and unmet healthcare need in a community-based sample of women who use cocaine, heroin or methamphetamine. Addict Behav 85 (2018) 94–99, 10.1016/j.addbeh.2018.05.013 [DOI] [PubMed] [Google Scholar]
- 17.Lanza ST and Rhoades BL, Latent class analysis: an alternative perspective on subgroup analysis in prevention and treatment. Prev Sci 14(2) (2013) 157–68, 10.1007/s11121-011-0201-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Conti AA, et al. , Chronic tobacco smoking and neuropsychological impairments: A systematic review and meta-analysis. Neurosci Biobehav Rev 96 (2019) 143–154, 10.1016/j.neubiorev.2018.11.017 [DOI] [PubMed] [Google Scholar]
- 19.Musk AW and de Klerk NH, History of tobacco and health. Respirology 8(3) (2003) 286–90, 10.1046/j.1440-1843.2003.00483.x [DOI] [PubMed] [Google Scholar]
- 20.Saha SP, et al. , Cigarette smoke and adverse health effects: An overview of research trends and future needs. Int J Angiol 16(3) (2007) 77–83, 10.1055/s-0031-1278254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.HHS, in The Health Consequences of Smoking-50 Years of Progress: A Report of the Surgeon General. 2014: Atlanta (GA). [Google Scholar]
- 22.Leung G, et al. , Behavioral disorders and diabetes-related outcomes among Massachusetts Medicare and Medicaid beneficiaries. Psychiatr Serv 62(6) (2011) 659–65, 10.1176/ps.62.6.pss6206_0659 [DOI] [PubMed] [Google Scholar]
- 23.Hongli Z, et al. , Joint effect of alcohol drinking and tobacco smoking on all-cause mortality and premature death in China: A cohort study. PLoS One 16(1) (2021) e0245670, 10.1371/journal.pone.0245670 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Wu LT, et al. , Substance use and mental diagnoses among adults with and without type 2 diabetes: Results from electronic health records data. Drug Alcohol Depend 156 (2015) 162–169, 10.1016/j.drugalcdep.2015.09.003 [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.
Supplementary Materials
Data Availability Statement
The data underlying this article will be shared on reasonable request to the corresponding author.