Study population and design

This prospective study explores predictors of self-reported continuous smoking abstinence in adult patients 6 weeks after hospital discharge. Electronic health records (EHRs) were used to measure potential predictors of smoking abstinence and included patient demographic factors (age, gender, race, insurance status), clinical characteristics such as length of hospital stay, Charlson Comorbidity score, other comorbidities such as alcohol or other substance use disorders, and exposure to the hospital’s TTP.

The study population includes 985 adult current smoking patients admitted to the hospital between March and December 2021 who were discharged home following their hospital stay and followed up 6 weeks later to reassess smoking status. To be included in our sample, patients had to have been hospitalized for at least 48 hours, be discharged to their home, have a working phone number, speak English, complete our Research Electronic Data Capture (REDCap) follow-up telephone survey, and acknowledge in the follow-up survey that they had been a current smoker at the time of admission to the hospital. There were 6000 patients referred to the TTP. Of the 6000 adult smokers, 985 completed our REDCap follow-up survey and acknowledged being current smokers at the time of admission to the hospital. During the study period, patients referred to the TTP were randomly assigned to one of two treatment options: 1) Enhanced care (EC), which consisted of a Tobacco Treatment Specialist (TTS) bedside consultation plus an automated post-discharge follow-up call designed to connect those smoking to the SCTQ; or 2) Basic care (BC), which consisted of the post-discharge follow-up call only. The bedside consultation consisted of an interview with a TTS to assess the patient’s smoking history, prior quitting attempts, and nicotine withdrawal symptoms. The TTS provided counseling to stop smoking, arranged for nicotine replacement therapy (NRT), and provided resources to aid in smoking cessation. Assignment to treatment conditions did not guarantee patients would actually receive the treatment offered (i.e. unavailable for bedside consultation). This study received ethics clearance through the Medical University of South Carolina IRB Committee (IRB# Protocol #107000) with a HIPAA Waiver of Authorization for Research, in which approval was granted for limited protected health information (PHI) in the EHR.

To conduct this secondary data analysis, we linked data from patients’ EHR, a REDCap telephone follow-up survey conducted six weeks after patients had been discharged back to their home, and TTP records data. The EHR data included date and time of admission, smoking status at time of admission, clinical covariates (i.e. length of stay, Charlson comorbidity index summary score, and comorbidities such as alcohol and other substance use disorders, chronic obstructive pulmonary disease [COPD], hypertension, neurological disorders, peripheral vascular diseases, among others), and demographics of our patients (i.e. age, gender, race, insurance status). The REDCap telephone follow-up survey six weeks post-discharge asked about the patient’s past smoking history (i.e. duration of smoking and age of first smoking purchase), smoking status since being discharged from the hospital, and the use of stop smoking medications since being discharged from the hospital. Lastly, TTP data included the intervention that the patient received (i.e. received bedside consultation, follow-up phone call after discharge).

Dependent variable - smoking abstinence

Self-reported smoking abstinence at the time of the follow-up call at 6 weeks was the primary dependent variable. Patients were asked if they had smoked a cigarette, even a puff, since being discharged. Patients who answered ‘Yes’ were considered to have relapsed back to smoking, and patients who answered ‘No’ were considered smoking abstinent since their discharge from the hospital.

Predictors of smoking abstinence

Predictors of smoking abstinence in this study fall into four categories: 1) patient demographics; 2) clinical factors related to a patient’s medical condition and hospital stay; 3) exposure to different elements of the TTP; and 4) self-reported use of stop smoking medications after discharge from the hospital. Each category of predictor variables is described below.

Patient demographics

The patient demographics we used were patients’ age, sex assigned at birth (male, female), race (White, Non-White), insurance status (insured, uninsured), and veteran status (veteran, not a veteran). Racial identities were dichotomized, given small cell sizes for some racial backgrounds (White, Non-White). We also used patients’ zip codes to assign them a social vulnerability score based on socioeconomic factors such as poverty, crowded housing, and lack of access to transportation. This score ranges from 0 to 1, with 0 being the least socially vulnerable and 1 being the most socially vulnerable²¹.

We also considered information about the patient’s smoking history obtained in the REDCap survey, which included information on duration of smoking and age of smoking initiation. The question on the duration of smoking asked patients to report how many years altogether they had smoked cigarettes. Patients were also asked to report at what age they first purchased cigarettes for themselves (<15, 15–17, 18–21, >21 years).

Clinical factors related to a patient’s medical condition and hospital stay

Clinical factors considered included the patient’s length of stay in the hospital, Charlson score (ranging from 0 to 13, with higher scores indicating higher risk of mortality)²², and various comorbidities determined from patients’ ICD-10 codes²³. These comorbidities included alcohol use disorder, cardiac arrhythmia, congestive heart failure, chronic obstructive pulmonary diseases (COPD), coagulopathy, deficiency anemia, depression, diabetes, drug use disorder, fluid/ electrolyte disorder, hypothyroidism, liver disease, obesity, other neurological disorders, pulmonary circulation disorder, peripheral vascular disorder, renal failure, tumor, vascular disorder, weight loss, and pneumonia.

Exposure to different elements of the TTP

To be consistent with our prior study evaluating the moderating impact of exposure to the TTP, we classified patients as exposed to different levels of the service (i.e. high, medium, and any exposure) versus no exposure^13,14. Patients who received both the bedside consultation and automated post-discharge follow-up call were considered to have had high exposure to the TTP. Patients who received exposure to either the bedside consultation or the post-discharge follow-up call, but not both, were considered to have had medium exposure. Patients who received either high or medium exposure were also considered to have had any exposure to TTP. Patients who received neither the consultation nor the follow-up call were considered to have had no exposure.

Use of smoking cessation medications after discharge from the hospital

In the REDCap telephone survey, we asked patients to report on their use of any smoking cessation medications (i.e. NRT such as gum, lozenge, inhaler, nasal spray, patch; varenicline/Chantix, and bupropion/Wellbutrin or Zyban) since they had been discharged from the hospital. A dichotomous smoking cessation medication use variable was created (used medication, did not use medication).

Statistical analysis

All analyses were conducted using SAS 9.4 (Cary, NC). Descriptive statistics included frequencies and percentages for categorical variables, and means and ranges for continuous variables. Logistic regression models were used to produce unadjusted bivariate odds ratios (ORs) and adjusted odds ratios (AORs). In the adjusted multivariate regression analyses of individual predictor variables, all other variables were considered as covariates in the adjusted models. A log transformation was used to address skewness for three variables: length of stay (days), the Charlson comorbidity score, and the social vulnerability index²⁴. Missing data on the predictor variables were relatively rare; therefore, we chose to exclude patients with missing data from our analyses rather than do imputation.

In an effort to isolate the impact of exposure to the different elements of the TTP on smoking abstinence, we used propensity score balancing to adjust for patient characteristics and clinical covariates^25,26. We used the inverse probability treatment weighted (IPTW) propensity score method for balancing exposure groups on available covariates. The propensity score balancing analysis produced weights used to balance the covariates between exposure groups. The distribution of covariates between exposure group conditions was assessed unadjusted and then reassessed using the inverse probability (propensity) weights generated by the propensity model. The success of the balancing was assessed by comparing the unadjusted group differences on each covariate against those differences observed after applying propensity score weights that were used to balance the groups. Differences were tested using chi-squared analyses for categorical variables and two-sample t-tests for continuous variables.

Covariates that were not successfully balanced with a p<0.25 were then evaluated in terms of their relationship with exposure to the TTP. Covariates that still had a p<0.25 in the individual assessments were included in a weighted logistic regression model to assess the association between smoking status and exposure to the TTP. Covariates that did not reach the p<0.25 cutoff were removed, and the model was re-fit in a stepwise fashion to create a parsimonious model. Supplementary file Tables 1a–1d show the effectiveness of the propensity score balancing of covariates for our four different TTP exposure condition comparisons. Only one model, comparing enhanced versus basic care conditions (Model D), required covariates to be included in the logistic model after propensity balancing weights were applied. The added covariates in the weighted logistic model are bolded in Supplementary file Table 1d.

Limitations

There are several limitations worth noting when interpreting current study results, including that smoking status was self-reported and not biochemically verified, which may have resulted in an overestimation of abstinence rates³¹. Indeed, self-reported smoking abstinence could introduce information bias (e.g. social-desirability bias) and misclassification bias. Additionally, residual confounding due to failure to adjust for unmeasured confounders could lead to over- or under-estimation of true associations. For example, the study did not collect information about important cessation predictors such as nicotine dependence, number of cigarettes smoked per day, past efforts to quit smoking, and the use of other nicotine-containing tobacco products (i.e. e-cigarettes, nicotine pouches). It may be the case that patients did make reductions in their tobacco use post-discharge, but this information was not captured. Also, while having standardized data on a patient’s post-discharge smoking status is a strength of this study because such data are rarely collected, the sample size available for measuring smoking abstinence was relatively modest compared to the large number of predictor variables examined, which can lead to unreliable interpretations of associations. Some cell sizes for diagnoses were small and may preclude us from reliably testing associations. Thus, we would urge readers to be cautious in interpreting the findings of reported associations.