Next Article in Journal
Differences between Very Highly Sensitized Kidney Transplant Recipients as Identified by Machine Learning Consensus Clustering
Previous Article in Journal
Melanoma: BRAFi Rechallenge
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Medicina
Volume 59
Issue 5
10.3390/medicina59050976
medicina-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Longitudinal Assessment of Multimorbidity Medication Patterns among Smokers in the COPDGene Cohort

by
Yisha Li
1,
Sarah J. Schmiege
2,
Heather Anderson
3,
Nicole E. Richmond
1,
Kendra A. Young
1,
John E. Hokanson
1,
Stephen I. Rennard
4,
Tessa L. Crume
1,
Erin Austin
5,
Katherine A. Pratte
6,
Rebecca Conway
1 and
Gregory L. Kinney
1,*
1
Department of Epidemiology, University of Colorado Anschutz Medical Campus, Aurora, CO 80045, USA
2
Department of Biostatistics and Informatics, University of Colorado Anschutz Medical Campus, Aurora, CO 80045, USA
3
Department of Clinical Pharmacy, University of Colorado Anschutz Medical Campus, Aurora, CO 80045, USA
4
Division of Pulmonary, Critical Care and Sleep Medicine, University of Nebraska Medical Center, Omaha, NE 68198, USA
5
Mathematical and Statistical Sciences, University of Colorado Denver, Denver, CO 80204, USA
6
Division of Biostatistics and Bioinformatics, National Jewish Health, Denver, CO 80206, USA
*
Author to whom correspondence should be addressed.
Medicina 2023, 59(5), 976; https://doi.org/10.3390/medicina59050976
Submission received: 15 March 2023 / Revised: 8 May 2023 / Accepted: 15 May 2023 / Published: 18 May 2023
(This article belongs to the Section Epidemiology & Public Health)
Browse Figures
Review Reports Versions Notes

Abstract

:
Background and objectives: Chronic obstructive pulmonary disease (COPD) is usually comorbid with other chronic diseases. We aimed to assess the multimorbidity medication patterns and explore if the patterns are similar for phase 1 (P1) and 5-year follow-up phase 2 (P2) in the COPDGene cohort. Materials and Methods: A total of 5564 out of 10,198 smokers from the COPDGene cohort who completed 2 visits, P1 and P2 visits, with complete medication use history were included in the study. We conducted latent class analysis (LCA) among the 27 categories of chronic disease medications, excluding COPD treatments and cancer medications at P1 and P2 separately. The best number of LCA classes was determined through both statistical fit and interpretation of the patterns. Results: We found four classes of medication patterns at both phases. LCA showed that both phases shared similar characteristics in their medication patterns: LC0: low medication; LC1: hypertension (HTN) or cardiovascular disease (CVD)+high cholesterol (Hychol) medication predominant; LC2: HTN/CVD+type 2 diabetes (T2D) +Hychol medication predominant; LC3: Hychol medication predominant. Conclusions: We found similar multimorbidity medication patterns among smokers at P1 and P2 in the COPDGene cohort, which provides an understanding of how multimorbidity medication clustered and how different chronic diseases combine in smokers.

1. Introduction

Cigarette smoking is a primary causative factor for a multitude of chronic diseases, including chronic obstructive pulmonary disease (COPD) [1,2]. In addition to COPD, people who smoke frequently have multimorbidity (the presence of multiple diseases or conditions), including heart disease, cancers and diabetes. Comorbidities increase the risk of poorer quality of life, higher disability life years lost, and earlier age at death [1]. Despite the high prevalence of comorbidities among those with COPD, longitudinal multimorbidity patterns are not well-understood [1,2]. In a cohort at high risk of COPD, typically observed morbidities include cardiovascular disease (CVD), diabetes, hyperlipidemia, and obesity [2]. Identifying diseases by diagnosis code is commonly used; however, this specificity is not always accessible in observational studies. In a previous study, we applied a novel approach using medication information as an indication of multimorbidity among smokers at high risk of COPD [3]. Previously, we identified 8 unique medication patterns among a total of 10,127 former and current smokers at phase (P1) in the COPDGene cohort [3]. These eight medication patterns independently predicted mortality and acute COPD exacerbations in the COPDGene cohort [3]. Since this study reported medication patterns at P1, a longitudinal assessment of medications of both P1 and phase 2 (P2) will be useful to describe medication patterns as an indication of multimorbidity patterns over time.
The aim of this study is to assess the prevalence and stability of multimorbidity treatment patterns at two study visits after accounting for P1 mortality and loss of follow-up in the COPDGene cohort. Our work is to provide a better understanding of both morbidity treatment and morbidity patterns to advance clinical insights into how different chronic diseases combine and whether multimorbidity medication (MM)patterns remain stable among smokers at high risk of COPD.

2. Materials and Methods

COPDGene is a multicenter observational study designed to identify genetic factors associated with COPD [4]. COPDGene enrolled 10,198 former and current smokers aged 45–80 years with a history of ≥10 pack-year smoking with and without COPD from initiation in P1 from 21 clinical centers across the United States (Figure 1) [4]. All participants were given consent before entering the study. COPDGene started P1 enrollment in 2008 and finished P1 enrollment in 2011, and five-year follow-up data were completed in P2 [5]. We used the COPDGene® to study P1 and P2 data. A total of 5564 participants who had both P1 and P2 medication and demographic information were included in our study cohort (Figure 1). Medication information was collected through self-report and confirmed through study personnel evaluation of participants’ bottles of medications, including both prescribed and over the count medication in each phase of their study visit. Medication exposure status was reported as “yes/no” to indicate if a specific medication was taken. Medication groups were further categorized based on expert opinion [3]. A total of 27 categories of medications were identified in both phases (Supplementary, Table S1. A list of self-reported morbidities in the COPDGene was published in previous literature [2]. The self-reported morbidities are dichotomous (yes/no) variables reported separately from medication use, which reflects whether participants self-reported having specific diseases.
Latent class analysis (LCA) using the 27 categories of medications was conducted to assess patterns of medication use. Identification of the best number of latent classes of medication patterns is achieved by testing multiple a priori number of classes and evaluating different model fits using goodness-of-fit indices. We initially evaluated two classes and increased the classes sequentially. According to the literature, there is some agreement that to select a final LCA model, (1) multiple fit statistics should be used or reported [6]; (2) the Bayesian information criterion (BIC) may be the most reliable fit statistic and should routinely be reported [6,7,8]; and (3) reasonable interpretability should be considered in choosing solution [6,7,9,10,11]. We included the comparison of Akaike information criterion (AIC), BIC, sample size adjusted BIC (SABIC), and entropy. The best model was finally determined by BIC, the likelihood-ratio test (LRT), and the overall interpretability of the model [12,13]. A lower value of AIC, BIC, and SABIC indicates a better model fit, and a higher value of entropy (≥0.8) shows increased classification precision [14,15]. After determining the optimal classes of LCA, all participants were classified into the class where they had the highest predicted probability of class membership, and the best class of membership for each individual was automatically selected through LCA. The label for each LCA class was determined by the observed item-response probabilities and labeled by names of dominating categories of medications. Item-response probability is considered high if the response probability is >0.7 and low if the response probability is <0.3. The naming of each latent class (LC) is based on the heterogeneity of medication categories in each LC, the names of predominant medication categories, and how the characteristics of each LC are distinguished from others [3,16,17]. We repeated the same LCA methods using the P2 data.
Latent transitional analysis (LTA), a longitudinal version of LCA, tested the longitudinal measurement invariance (likelihood-ratio test, LRT: p > 0.05) of item-response probabilities between P1 and P2 [18]. If the longitudinal measurement invariance in the LTA cannot be achieved (LRT, p < 0.05), the phase-specific LCA is used to report patterns over time. Either the LTA or the phase-specific LCA provided us with results of how individuals moved from one pattern to a different pattern or remained in the same pattern in each phase. We summarized the movement of medication patterns among individuals between phases to illustrate how MM patterns changed over time.
MM patterns with the participants’ self-reported comorbidities in each phase were compared through logistic regression to determine the utility of medication patterns serving as a proxy for diagnosis.
All statistical analyses were performed using SAS Version 9.4 (SAS Institute Inc.: Cary, NC, USA), and a p-value < 0.05 is considered statistically significant. LCA figures were created through the R package ggplot2, version 3.6.1 [19].

3. Results

The average age of participants was 60 years in P1; 50% of the participants were female, and 28% of participants identified as Non-Hispanic Black or African American (Table 1). A total of 47% of the participants reported being current smokers in P1, and the rest were former smokers (Table 1). The average body mass index (BMI) was 29 in P1.
The comparison of LCA fit statistics of the 2-class solution through the 8-class solution based on 27 classes of medications in P1 and P2 separately is shown in Table 2. BIC selected the 5-class solution and the 6-class solution in P1 and P2, respectively; however, the results of LCA were not interpretable after the 4-class solution in either phase. The BIC in the 3-class solution and 4-class solution were the same in P1, but an LRT showed the 4-class solution significantly improved model fit compared to the 3-class solution (G2Δ:241.39, dfΔ:28, p < 0.0001). The 5-class solution has the same BIC as the 6-class solution in P2, but the entropy is only 0.66 (Table 2). In addition, the 5-class solution model did not show as distinguished characteristics as the 4-class solution model in P2. Therefore, a 4-class solution model was also selected for P2.
LTA was applied for the 4-class solution models for both phases. However, we did not achieve measurement invariance because the non-constrained model showed a significantly better fit through LRT in both phases (LRT: p < 0.0001, Supplementary, Table S2). Therefore, results of the medication patterns are reported for the 4-class solution LCA model for each phase separately.

3.1. Characteristics of LCA Patterns at P1

The first of the four latent classes for P1 (denoted LC0_p1) included 3138 (56.40%) participants. Participants in the LC0_p1 had a low probability of all medication use (probability of each medication <6%, Figure 2), and the medications used in this group did not share the same clustering with other LCA classes. LC0_p1 was labeled as “Low medication”.
The second of the four latent classes for P1 (LC1_p1) included 690 (12.40%) participants. LC1_p1 had a high probability of angiotensin-converting-enzyme inhibitors (ACEi, 97.68%) use and a moderate probability of antiplatelet (40.52%), diuretic (41.31%) and statin (55.81%) use (Figure 2). Both ACEis and diuretics are medications primarily treating hypertension (HTN) [20]. Antiplatelets primarily decrease the risk of stroke, heart attack or other CVD [21]. Statins primarily reduce low-density lipoprotein and treat high cholesterol (Hychol) [22,23]. LC1_p1 was labeled as “HTN/CVD+ Hychol predominant”.
The third of the four latent classes for P1 (LC2_p1) included 255 (4.6%) participants. LC2_p1 had a high probability of biguanides (72.44%) and statin (73.20%) use, moderate probability of ACEi (45.45%), antiplatelet (46.50%) and sulfonylureas (40.34%) use (Figure 2). Biguanides and sulfonylureas are primarily used to treat T2D [24,25]. LC2_p1 was labeled as “HTN/CVD+T2D+Hychol predominant”.
The last latent classes for P1 (LC3_p1) included 1481 (26.6%) participants. LC3_p1 had a moderate probability of statin (47.21%) use (Figure 2). LC3_p1 was labeled as “Hychol predominant”.

3.2. Characteristics of LCA Patterns at P2

The first of the four latent classes for P2 (LC0_p2) included 2386 (42.90%) participants. LC0_p2 had a low probability of all medication use (probability of each medication <6%, Figure 2), and the medications used in this group did not share the same clustering with other LCA models at P2. LC0_p2 was labeled as “Low medication”.
The second of the four latent classes for P2 (LC1_p2) included 485 (8.70%) participants. LC1_p2 had a moderate probability of ACEi (41.15%) and antiplatelet (66.70%) (Figure 2). LC1_p2 had a high probability of beta blocker (98.17%) and statin (79.06%) use (Figure 2). ACEi and Beta blocker are medications that primarily treat HTN or CVD [20,26]. LC1_p2 was labeled as “HTN/CVD+Hychol predominant”.
The third of the four latent classes for P2 (LC2_p2) included 450 (8.10%) participants. LC2_p2 had a moderate probability of ACEi (51.69%) and antiplatelet (40.46%) use (Figure 2). LC2_p2 had a high probability of biguanides (74.11%) and statin (69.83%) use (Figure 2). LC2_p2 was labeled as “HTN/CVD+T2D+ Hychol predominant”.
The last latent class for P2 (LC3_p2) included 2,243 (40.30%) participants. LC3_p2 had a moderate probability of Statin (50.43%) use (Figure 2). LC3_p2 was labeled as “Hychol predominant”.
The numbers of individuals that changed medication patterns from P1 to P2 are shown in Table 3. Most participants remained in the same patterns in LC0 (low medication, 63.64%), LC2 (HTN/CVD + T2D + Hychol predominant, 52.55%) and LC3 (Hychol predominant, 62.19%) from P1 to P2. A total of 53.62% of participants in LC1 (HTN/CVD + Hychol predominant) in P1 transitioned to LC3 (Hychol predominant) in P2 (Table 3).
MM patterns are consistent with their primarily treated self-report comorbidity patterns (Supplementary, Table S3). There was a total of 1148 participants that died in the 5-year follow-up (P1), and 1134 of them had baseline medication information. We included their P1 medication patterns profile and death rate based on our previous published outcomes (Supplementary, Table S4) [3].
Figure 2 shows P1 and P2 medication patterns among participants who completed both phases: red color shows a higher probability (close to 1), and yellow color shows a lower probability (close to 0) of having a specific category of medication. X-axis includes labels of each medication pattern in phase 1 (left side) and phase 2 (right side) separately. Labels are explained as follows: LC0_p1 had a low probability of all medication use (probability of each medication <6%). LC1_p1 selected angiotensin-converting-enzyme inhibitors (ACEi, 97.68%), antiplatelet (40.52%), diuretics (41.31%) and statin (55.81%). LC2_p1 selected biguanides (72.44%), statin (73.20%), ACEi (45.45%), antiplatelet (46.50%) and sulfonylureas (40.34%). LC3_p1 selected statin (47.21%). LC0_p2 had a low probability of all medication use (probability of each medication <6%. LC1_p2 selected ACEi (41.15%), antiplatelet (66.70%), beta blocker (98.17%) and statin (79.06%). LC2_p2 selected ACEi (51.69%), antiplatelet (40.46%), biguanides (74.11%) and statin (69.83%). LC3_p2 selected statin (50.43%).

4. Discussion

We found consistent MM patterns based on a longitudinal assessment of 5564 current or former smokers who completed both P1 and P2 in the COPDGene cohort (Figure 2). The importance of looking at both P1 and P2 medication is to inform healthcare givers regarding the patterns of multiple medication use and long-term chronic conditions among smokers. Our work is innovative because our longitudinal medication patterns demonstrated that some chronic diseases progress concurrently, and our treatment patterns provide insights into examining and treating human diseases as an integrated paradigm. Our study has several strengths. To our knowledge, our study is the first study exploring longitudinal MM treatment patterns in the COPDGene cohort. We also applied an LTA approach to test whether the patterns at P1 and P2 are concordant from a statistical perspective. We identified the same four latent classes representing patterns of medication use at P1 and P2 in the COPDGene cohort: LC0: low medication, LC1: HTN/CVD + Hychol predominant, LC2: HTN/CVD + T2D + Hychol predominant, LC3: Hychol predominant. Similar medication patterns indicated that this cohort has consistent chronic diseases of HTN, CVD, T2D and hyperlipidemia. This is consistent with the report of consequences in smokers that heart disease, diabetes, stroke, and COPD are the most prevalent diseases among smokers in the past 50 years [1].
We did not include COPD medication in our analysis because participants’ pulmonary function was evaluated through spirometry and CT scan and can be identified through the global initiative of COPD status in Table 1. Even though the medication patterns showed similarity, we did not achieve measurement invariance in LTA. The reason we did not achieve measurement invariance is that some of the parameter estimates of the medication category were not the same in P1 and P2. For example, LC2_p1 and LC2_p2 were both labeled as HTN/CVD + T2D + Hychol medication; however, LC1_p1 selected both biguanides and sulfonylureas as potential T2D treatment, but LC2_p2 only selected biguanides. We found consistent MM patterns at a population level in the COPDGene cohort in both P1 and P2; however, the MM structure may not remain the same in each phase at an individual level. Individual treatment patterns could fluctuate over time because of changes in the number of diseases or changes in treatments or treatment regimens.
Our study showed there were changes in medication patterns at the individual level between groups from P1 to P2 (Table 3). Most participants in LC0, LC2 and LC3 remained in the same LCA class (Table 3). LC1 group had the most transition (53.62%) from LC1_p1 (HTN/CVD + Hychol predominant) to LC3_p2 (Hychol predominant) (Table 3). It is possible these people still had HTN or CVD in P2 and transitioned to less prevalent HTN/CVD treatment due to changes in their prescriptions, which are not captured by LCA because of the low frequency of use. It is also possible that some people started to manage HTN/CVD through lifestyle modification such as cessation of smoking, changing diet or exercise [27]. The numbers of medication patterns discovered in this work were different from our preliminary findings of 8 medication patterns in earlier published work, but this result was from data based on a total of 10,127 participants in P1 [3]. Subsequently, the difference is because our current study was based on 5564 participants (survivors of the 10,127 participants) who completed both phases, so there will be some changes in both numbers and characteristics of the medication patterns contributed either due to mortality or loss of follow-up. The differences in demographics in P1 among participants in our study cohort (completed both P1 and P2) and participants excluded are described in Supplementary, Table S5.
Even though the numbers of medication patterns were different from our previous study, we were still capturing most of the morbidities we found previously, including HTN, T2D, CVD and hyperlipidemia [3]. There was increased use of beta blocker in LC1 at P2 compared to P1 (Figure 2). This might be explained by the recommendations from both the American Heart Association and European Society of Cardiology that beta blocker and ACEi should be applied to all people who had heart failure with reduced ejection fraction to prevent symptomatic heart failure and reduce hospitalization rate and mortality [28,29,30]. Beta blocker was also a first-line treatment recommended for people who coexist with atrial fibrillation to control ventricular rate [28,29,30].
Our study had several limitations. First, there were some new medications included in P2 for diabetic treatment, which were not available for P1 because they were not introduced as a new treatment available in the market between the two phases. For example, Glucagon-like peptide 1 inhibitors were included in other diabetic medications in P2 but not in P1. However, this difference did not impact our study outcome because only two participants used this medication in P2, and other diabetic medications were not selected in the LCA model in either phase. Second, we did not collect information regarding patients’ medication dosage and adherence; thus, we cannot confirm whether participants used a specific medication. It is also likely to have report bias because patients might forget to bring their medication bottles to their visits which results in over-reporting or under-reporting their medication. Third, as we discussed in our previous paper, the same medication can be applied to treat different diseases, but we only make inferences about the primary diseases treated by medications [3]. For example, we cannot determine whether people were using HTN medications to treat chronic kidney disease (CKD) or HTN. The interaction between CKD and HTN is complex, but resistant HTN is commonly seen in patients with CKD [31]. Fourth, our LCA model indicates medication patterns but does not reflect every single medication used by each individual. However, the probability of using each category of medication within each pattern can be found in our LCA item-response probability. Fifth, it is possible that participants received more novel therapeutic interventions, lifestyle adjustments or over-the-counter medications to treat their diseases that were not included in our model because the information is not available in our data. However, it is an informative approach to capture medication patterns as a proxy of multimorbidity at a population level. Sixth, our study did not use validated comorbidity evaluation tools like the Charlson comorbidity index (CCI) score to confirm our approach because we do not have access to the diagnosis code for CCI calculation [32]; however, the predominant diseases captured through our approach were consistent with previously published literature [1]. Understanding medication patterns is also a useful tool for making inferences about disease patterns through treatment decisions when accurate diagnosis information is absent. Seventh, our study did not compare the difference in medication use or medication patterns between current and former smokers, which will be addressed in our future work. Lastly, our study is limited by survivor bias because we only included participants who completed both P1 and P2 with 5-year survival. Therefore, the generalization of these MM patterns applies to relatively healthier smokers. Future studies should reproduce MM patterns using other cohorts with smokers.

5. Conclusions

We found consistent MM patterns in each phase among 5564 participants who completed both P1 and P2 in the COPDGene study. Despite known changes in medication use guidelines and transitions of treatments, our approach showed there were similar MM patterns in smokers in two consecutive phases. Because chronic diseases are not curable even after being treated, these MM patterns are indications that captured consistent chronic disease patterns of HTN, CVD, T2D and hyperlipidemia among current and former smokers at a population level. Our finding provides an understanding of both morbidity treatment and morbidity patterns to advance clinical insights into how different chronic diseases combine in smokers.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/medicina59050976/s1, Table S1 A total of 27 classes medication classification and generic names shared in both P1 and P2, Table S2 Comparison of constrained and non-constrained latent transition model; Table S3 Comparison of medication patterns to self-report morbidities at P1 and P2 separately; Table S4 Death rate based on previous published P1 medication patterns; Table S5 Comparison of P1 demographics between included and excluded participants.

Author Contributions

Conceptualization, Y.L., S.J.S., H.A. and G.L.K.; Methodology, Y.L., S.J.S., H.A., N.E.R., K.A.Y., J.E.H., S.I.R., T.L.C., E.A., K.A.P., R.C. and G.L.K.; Software, Y.L. and S.J.S.; Validation: Y.L., S.J.S., H.A., N.E.R., K.A.Y., J.E.H., S.I.R., T.L.C., E.A., K.A.P., R.C. and G.L.K.; Formal analysis, Y.L. and S.J.S.; Investigation, Y.L., S.J.S., H.A., N.E.R., K.A.Y., J.E.H., S.I.R., T.L.C., E.A., K.A.P., R.C. and G.L.K.; Formal analysis, Y.L. and S.J.S.; Resources, J.E.H., K.A.Y. and G.L.K.; Data curation, J.E.H., K.A.Y. and G.L.K.; Writing—original draft preparation, Y.L.; Writing—review and editing, Y.L., S.J.S., H.A., N.E.R., K.A.Y., J.E.H., S.I.R., T.L.C., E.A., K.A.P., R.C. and G.L.K.; Visualization, Y.L.; Supervision, G.L.K.; Project administration, Y.L.; Funding acquisition, J.E.H., K.A.Y. and G.L.K. All authors have read and agreed to the published version of the manuscript.

Funding

The work was supported by award number U01 HL089897 and award number U01 HL089856 from the National Heart, Lung, and Blood Institute.

Institutional Review Board Statement

Protocols for all phases (HS-2778) and data evaluation in the COPDGene cohort followed the priciples of the Declaration of Helsinki and was approved by the COPDGene committee.

Informed Consent Statement

All study participants were informed about the purpose of the study. Confirmation of consent form was received by pariticpants.

Data Availability Statement

Data supporting this manuscript is available to all COPDGene investigators. Data are available upon request through https://dccweb.njhealth.org/sec/COPDGene/MainPage.cfm (1 May 2023).

Acknowledgments

We thank COPDGene investigators who helped acquiring the data for this study. COPDGene is supported by Award Number U01 HL089897 and Award Number U01 HL089856 from the National Heart, Lung, and Blood Institute. The central institutional reviewer board (IRB) number for COPDGene phase 1, phase 2 and phase 3 is HS-2778. COPDGene is also supported by the COPD Foundation through contributions made to an Industry Advisory Board that has included AstraZeneca, Bayer Pharmaceuticals, Boehringer-Ingelheim, Genentech, GlaxoSmithKline, Novartis, Pfizer, and Sunovion. We thank the Colorado School of Public Health Laboratory for Analytical and Computational Epidemiology (LACE) for their support.

Conflicts of Interest

In the last three years, Stephen I Rennard has received fees for consulting from GlaxoSmithKline, Verona, Boehringer Ingelheim, Novo Ventures and Sanofi. He is the Medical Director for the Alpha 1 Foundation Therapeutic Development Network. From 2015–2019 he was an employee of AstraZeneca and received shares as part of his compensation. Yisha Li is an employee of Vertex Pharmaceuticals and has received compensation from the company since April 2023.

References

  1. U.S. Department of Health and Human Services. The Health Consequences of Smoking—50 Years of Progress. A Report of the Surgeon General; U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health: Atlanta, GA, USA, 2014.
  2. Putcha, N.; Puhan, M.A.; Drummond, M.B.; Han, M.K.; Regan, E.A.; Hanania, N.A.; Martinez, C.H.; Foreman, M.; Bhatt, S.P.; Make, B.; et al. A simplified score to quantify comorbidity in COPD. PLoS ONE 2014, 9, e114438. [Google Scholar] [CrossRef] [PubMed]
  3. Li, Y.; Ragland, M.; Austin, E.; Young, K.; Pratte, K.; Hokanson, J.E.; Beaty, T.H.; Regan, E.A.; Rennard, S.I.; Wern, C.; et al. Co-Morbidity Patterns Identified Using Latent Class Analysis of Medications Predict All-Cause Mortality Independent of Other Known Risk Factors: The COPDGene(®) Study. Clin. Epidemiol. 2020, 12, 1171–1181. [Google Scholar] [CrossRef]
  4. Regan, E.A.; Hokanson, J.E.; Murphy, J.R.; Make, B.; Lynch, D.A.; Beaty, T.H.; Curran-Everett, D.; Silverman, E.K.; Crapo, J.D. Genetic epidemiology of COPD (COPDGene) study design. COPD J. Chronic Obstr. Pulm. Dis. 2010, 7, 32–43. [Google Scholar] [CrossRef] [PubMed]
  5. Ragland, M.F.; Strand, M.; Baraghoshi, D.; Young, K.A.; Kinney, G.L.; Austin, E.; Hokanson, J.E.; Regan, E.; Crapo, J.; Silverman, E.K.; et al. 10-Year Follow-Up of Lung Function, Respiratory Symptoms, and Functional Capacity in the COPDGene Study. Ann. Am. Thorac. Soc. 2022, 19, 381–388. [Google Scholar] [CrossRef]
  6. Weller, B.E.; Bowen, N.K.; Faubert, S.J. Latent class analysis: A guide to best practice. J. Black Psychol. 2020, 46, 287–311. [Google Scholar] [CrossRef]
  7. Nylund, K.L.; Asparouhov, T.; Muthén, B.O. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Struct. Equ. Model. A Multidiscip. J. 2007, 14, 535–569. [Google Scholar] [CrossRef]
  8. Vermunt, J.K. Latent class analysis of complex sample survey data: Application to dietary data. J. Am. Stat. Assoc. 2002, 97, 736–737. [Google Scholar]
  9. Muthén, B.; Muthén, L.K. Integrating person-centered and variable-centered analyses: Growth mixture modeling with latent trajectory classes. Alcohol. Clin. Exp. Res. 2000, 24, 882–891. [Google Scholar] [CrossRef]
  10. Shanahan, L.; Copeland, W.E.; Worthman, C.M.; Erkanli, A.; Angold, A.; Costello, E.J. Sex-differentiated changes in C-reactive protein from ages 9 to 21: The contributions of BMI and physical/sexual maturation. Psychoneuroendocrinology 2013, 38, 2209–2217. [Google Scholar] [CrossRef]
  11. Stringaris, A.; Maughan, B.; Copeland, W.S.; Costello, E.J.; Angold, A. Irritable mood as a symptom of depression in youth: Prevalence, developmental, and clinical correlates in the Great Smoky Mountains Study. J. Am. Acad. Child Adolesc. Psychiatry 2013, 52, 831–840. [Google Scholar] [CrossRef]
  12. Lo, Y.; Mendell, N.; Rubin, D. Testing the number of components in a normal mixture. Biometrika 2001, 88, 767–778. [Google Scholar] [CrossRef]
  13. Masyn, K.E. Latent class analysis and finite mixture modeling. In The Oxford Handbook of Quantitative Methods: Statistical Analysis; Little, T.D., Ed.; Oxford University Press: New York, NY, USA, 2013. [Google Scholar]
  14. Lanza, S.T.; Collins, L.M.; Lemmon, D.R.; Schafer, J.L. PROC LCA: A SAS Procedure for Latent Class Analysis. Struct. Equ. Model. Multidiscip. J. 2007, 14, 671–694. [Google Scholar] [CrossRef]
  15. Celeux, G.; Soromenho, G. An entropy criterion for assessing the number of clusters in a mixture model. J. Classif. 1996, 13, 195–212. [Google Scholar] [CrossRef]
  16. Linda, M.; Collins, S.T.L. Latent Class and Latent Transition Analysis with Applications in the Social, Behavioral, and Health Sciences; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2010. [Google Scholar]
  17. Liu, Y.; Elliott, A.L.; Serdarevic, M.; Leeman, R.F.; Cottler, L.B. A latent class analysis of the past-30-day substance use patterns among lifetime cocaine users: Findings from a community sample in North Central Florida. Addict. Behav. Rep. 2019, 9, 100170. [Google Scholar] [CrossRef]
  18. Ryoo, J.H.; Wang, C.; Swearer, S.M.; Hull, M.; Shi, D. Longitudinal Model Building Using Latent Transition Analysis: An Example Using School Bullying Data. Front. Psychol. 2018, 9, 675. [Google Scholar] [CrossRef]
  19. Wickham, H. ggplot2: Elegant Graphics for Data Analysis; Springer: New York, NY, USA, 2016. [Google Scholar]
  20. Morell, S.M.; Heine, G.H.; Fassnacht, M. Update on treatment resistant hypertension and secondary hypertension. Dtsch. Med. Wochenschr. 2021, 146, 742–746. [Google Scholar] [CrossRef]
  21. Mauri, L.; Kereiakes, D.J.; Normand, S.L.; Wiviott, S.D.; Cohen, D.J.; Holmes, D.R.; Bangalore, S.; Cutlip, D.E.; Pencina, M.; Massaro, J.M. Rationale and design of the dual antiplatelet therapy study, a prospective, multicenter, randomized, double-blind trial to assess the effectiveness and safety of 12 versus 30 months of dual antiplatelet therapy in subjects undergoing percutaneous coronary intervention with either drug-eluting stent or bare metal stent placement for the treatment of coronary artery lesions. Am. Heart J. 2010, 160, 1035–1041.e1. [Google Scholar] [CrossRef]
  22. Grundy, S.M.; Feingold, K.R. Guidelines for the Management of High Blood Cholesterol. In Endotext; Feingold, K.R., Anawalt, B., Blackman, M.R., Boyce, A., Chrousos, G., Corpas, E., de Herder, W.W., Dhatariya, K., Hofland, J., Dungan, K., et al., Eds.; MDText.com, Inc. Copyright © 2000–2023; MDText.com, Inc.: South Dartmouth, MA, USA, 2000. [Google Scholar]
  23. De Boer, L.M.; Oorthuys, A.O.J.; Wiegman, A.; Langendam, M.W.; Kroon, J.; Spijker, R.; Zwinderman, A.H.; Hutten, B.A. Statin therapy and lipoprotein(a) levels: A systematic review and meta-analysis. Eur. J. Prev. Cardiol. 2022, 29, 779–792. [Google Scholar] [CrossRef]
  24. Tomlinson, B.; Patil, N.G.; Fok, M.; Chan, P.; Lam, C.W.K. The role of sulfonylureas in the treatment of type 2 diabetes. Expert Opin. Pharmacother. 2022, 23, 387–403. [Google Scholar] [CrossRef]
  25. Krentz, A.J.; Bailey, C.J. Oral antidiabetic agents: Current role in type 2 diabetes mellitus. Drugs 2005, 65, 385–411. [Google Scholar] [CrossRef]
  26. Mann, S.J. Redefining beta-blocker use in hypertension: Selecting the right beta-blocker and the right patient. J. Am. Soc. Hypertens. 2017, 11, 54–65. [Google Scholar] [CrossRef] [PubMed]
  27. Samadian, F.; Dalili, N.; Jamalian, A. Lifestyle Modifications to Prevent and Control Hypertension. Iran. J. Kidney Dis. 2016, 10, 237–263. [Google Scholar] [PubMed]
  28. McMurray, J.J.; Adamopoulos, S.; Anker, S.D.; Auricchio, A.; Böhm, M.; Dickstein, K.; Falk, V.; Filippatos, G.; Fonseca, C.; Gomez-Sanchez, M.A.; et al. ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure 2012: The Task Force for the Diagnosis and Treatment of Acute and Chronic Heart Failure 2012 of the European Society of Cardiology. Developed in collaboration with the Heart Failure Association (HFA) of the ESC. Eur. Heart J. 2012, 33, 1787–1847. [Google Scholar] [CrossRef] [PubMed]
  29. Yancy, C.W.; Jessup, M.; Bozkurt, B.; Butler, J.; Casey, D.E., Jr.; Drazner, M.H.; Fonarow, G.C.; Geraci, S.A.; Horwich, T.; Januzzi, J.L.; et al. 2013 ACCF/AHA guideline for the management of heart failure: A report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines. Circulation 2013, 128, e240–e327. [Google Scholar] [CrossRef]
  30. Dézsi, C.A.; Szentes, V. The Real Role of β-Blockers in Daily Cardiovascular Therapy. Am. J. Cardiovasc. Drugs 2017, 17, 361–373. [Google Scholar] [CrossRef]
  31. Hamrahian, S.M.; Falkner, B. Hypertension in Chronic Kidney Disease. Adv. Exp. Med. Biol. 2017, 956, 307–325. [Google Scholar] [CrossRef]
  32. Charlson, M.E.; Carrozzino, D.; Guidi, J.; Patierno, C. Charlson Comorbidity Index: A Critical Review of Clinimetric Properties. Psychother. Psychosom. 2022, 91, 8–35. [Google Scholar] [CrossRef]
Medicina 59 00976 g001 550
Figure 1. Inclusion of study cohort: 10,127 out of 10,198 smokers had P1 medication data. A total of 1134 out of 10,127 participants died before P2, and 2811 participants had a loss of follow-up at P2. A total of 6235 participants had P2 demographic information, and 5564 participants had P2 medication information.
Figure 1. Inclusion of study cohort: 10,127 out of 10,198 smokers had P1 medication data. A total of 1134 out of 10,127 participants died before P2, and 2811 participants had a loss of follow-up at P2. A total of 6235 participants had P2 demographic information, and 5564 participants had P2 medication information.
Medicina 59 00976 g001
Medicina 59 00976 g002 550
Figure 2. Medication patterns based on 27 classes of medications at P1 and P2.
Figure 2. Medication patterns based on 27 classes of medications at P1 and P2.
Medicina 59 00976 g002
Table
Table 1. Demographics of participants who completed both P1 and P2.
Table 1. Demographics of participants who completed both P1 and P2.
P1 (n = 5564)P2 (n = 5564)
n (%)Mean (SD)n (%)Mean (SD)
Age 60.09 (8.68) 65.65 (8.66)
Female 2780 (49.96) 2780 (49.96)
African American 1584 (28.47) 1584 (28.47)
Smoking statusFormer smoker2957 (53.15) 3479 (62.53)
Current smoker2607 (46.85) 2085 (37.62)
Smoking pack-years 42.89 (23.62) 44.35 (23.99)
BMI 29.09 (6.11) 29.00 (6.39)
COPD GOLD PRISm689 (12.38) 665 (12.43)
GOLD 02541 (45.67) 2265 (42.32)
GOLD 1492 (8.84) 511 (9.55)
GOLD 21122 (20.17) 1092 (20.40)
GOLD 3570 (10.24) 575 (10.74)
GOLD 4150 (2.70) 244 (4.56)
BMI: body mass index, COPD GOLD: global initiative for chronic obstructive pulmonary disease, P1: phase 1, P2: phase 2 (5-year follow-up), PRISm: preserved ratio impaired spirometry, SD: standard deviation. The total number is not 5564 because of missing data.
Table
Table 2. Latent class analysis statistics of medication patterns in P1 and P2.
Table 2. Latent class analysis statistics of medication patterns in P1 and P2.
G2AICBICCAICABICEntropy
P1
2-solution7501.197611.197975.518030.517800.740.69
3-solution6936.087102.087651.887734.887388.130.70
4-solution6694.696916.697651.867762.867299.140.76
5-solution6360.606638.607559.347698.347117.640.71
6-solution6219.146553.147659.367826.367128.680.66
7-solution6116.416506.417798.107993.107178.450.66
8-solution5985.846431.847909.008132.007200.380.71
P2
2-solution11,622.0811,732.0812,096.4012,151.4011,921.630.68
3-solution11,027.3511,193.3511,743.1511,826.1511,479.400.57
4-solution10,600.4710,822.4711,557.7411,668.7411,205.020.69
5-solution10,232.2110,510.2111,430.9611,569.9610,989.260.66
6-solution9988.9510,322.9511,429.1711,596.1710,898.500.70
7-solution9826.1110,216.1111,507.8111,702.8110,888.160.73
8-solution9677.3110,123.3111,600.4711,823.4710,891.850.70
AIC: Akaike Information Criterion, BIC: Bayesian information criterion, SABIC: sample size adjusted BIC, CAIC: consistent AIC, P1: phase 1, P2: phase 2 (5-year follow-up).
Table
Table 3. Numbers of individuals in each medication patterns in P1 and P2.
Table 3. Numbers of individuals in each medication patterns in P1 and P2.
P2 Low Medication (n = 2386)P2 HTN/CVD + Hychol Predominant (n = 485)P2 HTN/CVD + T2D + Hychol Predominant
(n = 450)		P2 Hychol Predominant (n = 2243)
P1 Low medication (n = 3138)1997 (63.64)95 (3.03)133 (4.24)913 (29.09)
P1 HTN/CVD + Hychol predominant
(n = 690)		80 (11.59)140 (20.29)100 (14.49)370 (53.62)
P1 HTN/CVD + T2D + Hychol predominant
(n = 255)		24 (9.41)58 (22.75)134 (52.55)39 (15.29)
P1 Hychol predominant
(n = 1481)		285 (19.24)192 (12.96)83 (5.60)921 (62.19)
LC: latent class, P1: phase 1, P2: phase 2 (5-year follow-up), HTN: hypertension, CVD: cardiovascular disease, Hychol: high cholesterol, T2D: type 2 diabetes.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Li, Y.; Schmiege, S.J.; Anderson, H.; Richmond, N.E.; Young, K.A.; Hokanson, J.E.; Rennard, S.I.; Crume, T.L.; Austin, E.; Pratte, K.A.; et al. Longitudinal Assessment of Multimorbidity Medication Patterns among Smokers in the COPDGene Cohort. Medicina 2023, 59, 976. https://doi.org/10.3390/medicina59050976

AMA Style

Li Y, Schmiege SJ, Anderson H, Richmond NE, Young KA, Hokanson JE, Rennard SI, Crume TL, Austin E, Pratte KA, et al. Longitudinal Assessment of Multimorbidity Medication Patterns among Smokers in the COPDGene Cohort. Medicina. 2023; 59(5):976. https://doi.org/10.3390/medicina59050976

Chicago/Turabian Style

Li, Yisha, Sarah J. Schmiege, Heather Anderson, Nicole E. Richmond, Kendra A. Young, John E. Hokanson, Stephen I. Rennard, Tessa L. Crume, Erin Austin, Katherine A. Pratte, and et al. 2023. "Longitudinal Assessment of Multimorbidity Medication Patterns among Smokers in the COPDGene Cohort" Medicina 59, no. 5: 976. https://doi.org/10.3390/medicina59050976

Article Metrics

Back to TopTop