The Impact of Insurance Gain and Discussions with Healthcare Providers on Quitting Smoking

Abstract

Tobacco use is the leading preventable cause of death in the United States. Analyzing the ability for different mechanisms to reduce smoking rates can provide healthcare systems with information to establish the most effective smoking cessation efforts. Health insurance provides individuals with direct mechanisms to curb smoking behavior, such as access to smoking cessation resources. Gaining insurance may additionally indirectly influence smoking cessation by altering risk perceptions. Behavioral economic theory suggests that gaining health insurance may reduce current smokers’ rate of discounting on the future, which could increase smoking cessation. This paper aimed to evaluate the impact of insurance status (i.e., gaining any private (n=681), gaining only public (n=647), or remaining uninsured (n=5,056)) as well as the impact of having a discussion with a healthcare provider about quitting smoking on smoking cessation among current adult smokers who were uninsured at the beginning of their data collection. Data for this study came from the Medical Expenditure Panel Survey 2003 to 2014 database. The study found that while individuals gaining public insurance was not statistically associated with smoking cessation, individuals who gained private insurance were more likely to stop smoking than individuals who remained uninsured (OR: 1.330; 95% CI: 1.019,1.737; p=0.036). Having a discussion with a healthcare provider about quitting smoking was not associated with smoking cessation. These findings indicate that gaining private insurance may impact smoking behavior through mechanisms other than direct access to physician services.

Introduction

Despite national tobacco smoking rates falling from 43% to 18% between 1965 and 2014, nearly 40 million adults in the United States are current cigarette smokers.^1,2 Cigarette smoking leads to nearly 450,000 deaths annually in the United States.³ In addition to the increased mortality and morbidity, tobacco use also causes substantial economic burden. It is estimated that cigarette smoking leads to $130 billion in direct medical costs and $150 billion in lost productivity annually because of illness and early death due to smoking.¹

There have been numerous programs and initiatives that have contributed to the decline in smoking rates over the past 50 years.⁴ For example, excise taxes on tobacco products, media restrictions on tobacco advertisement, and increases in availability and access to smoking cessation programs have all successfully lowered smoking rates.^4–7 Yet, cigarette smoking remains the leading preventable cause of death in the United States.¹ Additional evaluations that determine mechanisms that may be able to further reduce smoking behaviors and increase smoking cessation can provide healthcare systems, public health practitioners, as well as clinicians with information to create effective smoking cessation programs. One potential mechanism for increasing an individual’s likelihood of smoking cessation is through gaining health insurance coverage.⁸

Improving access to high quality healthcare through increases in insurance coverage is a national priority. While a number of studies have evaluated the impact of health insurance on healthcare utilization and health outcomes,⁹ there exist relatively few studies evaluating the impact of gaining health insurance on health behaviors, such as cigarette smoking. Gaining health insurance may provide individuals with direct access to the healthcare system, such as increased access to primary care visits and smoking cessation resources. Acquisition of health insurance may additionally increase smoking cessation indirectly by altering an individual’s risk perception.

There are a number of psychological and cognitive constraints that may influence an individual’s choice to smoke that can be evaluated based on behavioral economic theory. Behavioral economics has been suggested as a tool to facilitate creating effective healthcare policies by understanding how humans make health-related decisions as well as how they respond to health and public health policies.^10,11

One behavioral economic mechanism for influencing smoking cessation is the potential for insurance to affect risk perception by altering an individual’s time preference. Time preference has been used to explain variation in individuals’ decisions to make unhealthy choices.¹² Time preference can be described as a preference for current consumption over future consumption.¹³ Individuals with high time preference place a greater value on current benefit so that they forgo choices that may have a higher future benefit. For example, an individual may make a myopic decision to spend money today to purchase goods while he or she could invest that money and buy a greater quantity of goods a year from today.

It has been suggested that time preference is a function of investing.¹⁴ Investments that increase an individual’s length of life or increase the value that an individual places on future consumption may reduce that individual’s time preference (i.e., make them more future oriented).¹⁵ Furthermore, increases in health insurance and access to medical care could increase an individual’s perceived longevity. Higher perceived longevity has been associated with healthier behaviors and utilization of preventive services.¹⁶ Stated another way, obtaining insurance should increase health capital and lower time preference. It could, therefore, be hypothesized that gaining insurance could lead to increased smoking cessation by reducing current smokers’ rate of discounting on the future.

To our knowledge, there exists only one evaluation of the impact of gaining insurance on an individual’s smoking cessation using a nationally representative adult sample.¹⁷ The study by Jerant et al.¹⁷ found that there was no impact of gaining insurance on smoking cessation. However, their study did not include the type of insurance (i.e., public or private) as a component in their analyses. Other studies indicate that individuals who gain insurance of different types may have different propensities of initiating as well as quitting smoking.^17–20 For example, one study found that men who gained Medicare had increases in unhealthy behaviors, including smoking, which differs from the findings in the study by Jerant et al.^17,19

In addition to evaluating the impact of gaining public versus private insurance, the current study includes a measure of an individual’s smoking cessation discussions with a healthcare provider. Provider counseling for smoking cessation has been found to be effective in improving smoking cessation rates.²¹ Using this variable in the analyses can control for an individual’s use of smoking cessation resources. This can provide information regarding if gaining insurance improves smoking cessation rates because of increased access to and utilization of smoking cessation resources, such discussions with healthcare providers about smoking cessation, or through other mechanisms, such as changes in time preference.

The purpose of this study was to analyze the impact of gaining insurance of different types as well as the impact of discussing smoking cessation with a provider on smoking cessation among previously uninsured current smokers of age 18 and older using data from a nationally representative sample from years 2003 through 2014 of the Medical Expenditure Panel Survey (MEPS).²² A better understanding of the relationships between insurance coverage, discussions with providers about quitting smoking, and the likelihood of quitting smoking can provide healthcare systems, clinicians, and public health officials with information about the underlying mechanisms behind an individual’s likelihood to quit smoking. It was expected that both gaining insurance and discussing smoking cessation with a healthcare provider would lead to more frequent smoking cessation.^8,21 Based on previous literature,^17,19 it was anticipated that individuals who gain different insurance types will have varying frequencies of smoking cessation.

Methods

Data

The data for this study came from the household component of the Medical Expenditure Panel Survey (MEPS), which is administered by telephone interviews and subsequently mailed self-administered supplemental questionnaires. The MEPS is administered by the Agency of Healthcare Research and Quality, and it is nationally and regionally representative of the non-institutionalized civilian United States population.²² The MEPS uses a stratified, multistage, area probability sampling process from individuals from the previous year’s National Health Interview Survey, and it oversamples specific subpopulations, including ethnic minorities and low-income individuals.

The MEPS household component collects data regarding healthcare utilization and expenditure, health status, as well as sources of payment and insurance coverage from a nationally representative sample of participants across the United States.²² This component of the survey uses an overlapping panel design with a new panel of sample households chosen each year. Data for each panel is collected in five rounds, and each household responds to interviews about every 5 months. Respondents can be linked across their individual interviews, which span over two years. This allows analyses to assess the impact of changes in individual characteristics on healthcare utilization and expenditure as well as on health-related outcomes. Unless otherwise specified, this study uses data from Rounds 2 and 4, which are the rounds in which smoking behavior questions are asked. In the remainder of this paper, “Year 1” will refer to data collected at Round 2 and “Year 2” will refer to data collected at Round 4 unless otherwise specified.

The data for this study came from years 2003 through 2014 (i.e., Panel 8 through Panel 18) of the publically available “Longitudinal Data Files” from MEPS (http://meps.ahrq.gov/mepsweb/data_stats/download_data_files.jsp). Each of these longitudinal data files links each individual’s responses across that individual’s interviews into one data file with person weights that adjust for the survey’s complex sampling design and nonresponse to provide population estimates. Survey stratum and primary sampling unit variables provided in the “Longitudinal Data Files” were used to account for the variance structure using multiple years of pooled data. Across the 11 panels, the point-in-time response rate varied from 58.0% to 68.9%, and the full-year response rate varied from 52.8% to 64.5%.²³

The University of Arkansas for Medical Sciences Institutional Review Board has reviewed and determined that this study is exempt from the IRB approval process.

Analytic sample

The sample population for this study included all individuals over the age of 19 at Year 2 who were uninsured and who were current smokers during Year 1 of their respective MEPS data collection. An age of 19 at Year 2 was used so that in Year 1, an individual was at least age 18, which is the age at which MEPS firsts asks smoking behavior questions. An individual was considered to be uninsured in Year 1 of the MEPS data collection if he or she had no insurance coverage of any type on the date of the Round 2 survey data collection. An individual was considered to be a current smoker if he or she answered “Yes” to the question “Do you currently smoke?” during Year 1 of his or her panel data.

The entire MEPS data for Panel 8 through Panel 18 include 179,208 individuals. Respondents were excluded if they were under the age of 19 at Year 2 (n=53,636) or if they had a missing age in both Years 1 and 2 (n=1,459) (Figure 1). Smoking status questions are taken from the MEPS Adult Self-Administered Questionnaire (SAQ). Among the remaining 124,113 individuals, individuals who did not respond to the SAQ in Rounds 2 and 4 were excluded (n=14,219). After additionally excluding individuals who were insured in Year 1 (n=81,726), 28,168 uninsured individuals remained in the study. Among those respondents, individuals were removed if they had a missing smoking status in Year 1 (n=626) or if they were not a current smoker during Year 1 (n=20,457). Among those 7,085 uninsured current smokers in Year 1, 139 individuals were excluded for having a missing smoking status in Year 2. Finally, individuals who didn’t answer the question regarding whether they had a discussion with a provider about smoking cessation were excluded (n=562). This resulted in an analytic sample equal to 6,384 individuals for the study.

Figure 1:

Inclusion/exclusion criteria and analytic sample size

Variables

Dependent variable

The dependent variable used in this study was a binary variable that indicated whether or not an individual stopped smoking. An individual was considered to have stopped smoking and given a value of 1 if he or she indicated to not be a current smoker during Year 2. An individual was considered to be a current smoker if he or she responded “yes” to the following question in the MEPS data: “Do you currently smoke?”. Individuals who were still current smokers in Year 2 were given a 0 value for the dependent variable.

Independent variables

This study included analyses for two separate independent variables of interest. The primary independent variable of interest was whether or not an individual gained insurance coverage between Year 1 and Year 2 of the survey data collection, and if a respondent did gain insurance, the type of insurance the individual gained was determined.

An individual was considered to have gained insurance coverage if he or she had any insurance coverage on the date of the Round 4 survey data collection. Other studies²⁴ have determined insurance status by examining the number of months of insurance coverage an individual had in Years 1 and 2. However, we were not interested in the insurance status for the arbitrary 12-month period in Year 1 and Year 2, but rather we are interested in the change in insurance status from uninsured to insured between the Round 2 and Round 4 interview dates.

Insurance status type was categorized as “Any Private” or “Only Public.” Previous studies vary with respect to assignment of an individual who has multiple insurance coverages.^24,25 For the purposes of this study, the approach that was used by a majority of the studies was chosen,^24,26 and an individual who had both private and public insurance coverage was categorized as “Any Private”. Although ideally different types of public insurance statuses (e.g., Medicaid or Medicare) could be analyzed separately, the relatively small sample size for insurance gains among these insurance types necessitated aggregating them into one “Only Public” category.

The second independent variable used in this study was whether or not an individual had a discussion with a healthcare provider about quitting smoking. Specifically, it was an individual’s response in his/her Year 1 MEPS Adult SAQ to the question “In the past 12 months did a doctor advise you to quit smoking?” that was used for this variable.

Statistical Analyses

Weighted percentages and chi square tests were calculated to detect differences in insurance statuses (i.e., Any Private, Only Public, or Remained Uninsured) among different demographic characteristics. Logistic regressions were conducted to assess associations in three separate models. Specifically, logistic regression analyses were performed to study 1) the impact of gaining private or public insurance on the likelihood of quitting smoking compared to remaining uninsured, 2) the impact of discussing smoking cessation with a provider on the likelihood of quitting smoking, 3) the impact of both primary independent variables (i.e., insurance status and smoking cessation discussion with a provider) on the likelihood of quitting smoking in a single model.

Each logistic regression was conducted with and without additional covariates. Covariates included individual-level factors based on the Andersen behavioral model and previous evaluations that used MEPS data^17,24,27 as well as the panel year to adjust for secular trends (Table 1). The individual-level covariates included gender, age category, race, education, annual family income as a percent of federal poverty level, marriage status, employment, region, health status, flu shoot received in the past year (as a proxy for other preventative behavior), and total healthcare spending in last 12 months. To avoid multicollinearity, the total healthcare spending variable was not used in logistic regressions that included the measure for discussion with a provider about smoking cessation. For covariates in which there were multiple data collection points, responses from the survey panel year for which the primary outcome measure was taken were used. Specifically, demographic responses from Round 4 (or Year 2 for variables with no round-level responses) were used. The flu shot measure was taken from Round 5.

Table 1.

Insurance status by demographic characteristics, among those uninsured current smokers of age 18 and older at Year 1, Medical Expenditure Panel Survey Longitudinal Data Files 2003–2004 through 2013–2014 (n=6,384)

	Remained Uninsured (N=5056)		Gained Insurance				Chi Sq p-value
			Any Private (N=681)		Only Public (N=647)
	N	Weighted % (95% CI)	N	Weighted % (95% CI)	N	Weighted % (95% CI)
Gender							<.0001
Male a	3102	81.0 (79.4, 82.7)	376	12.4 (10.8, 14.0)	254	6.6 (5.5, 7.7)
Female	1954	72.7 (70.6, 74.8)	305	14.5 (12.7, 16.2)	393	12.8 (11.4, 14.2)
Age							<.0001
19–29 a	1561	77.7 (75.3, 80.1)	220	13.2 (11.1, 15.2)	210	9.2 (7.7, 10.7)
30–39	1189	77.4 (74.2, 80.6)	169	14.1 (11.4, 16.8)	145	8.5 (6.7, 10.3)
40–54	1766	80.0 (77.7, 82.3)	223	13.3 (11.2,15.4)	165	6.7 (5.4, 8.0)
55>	540	72.2 (68.0, 76.4)	69	11.1 (8.1, 14.1)	127	16.7 (13.4, 20.0)
Race							<.0001
White a	2169	75.5 (73.5, 77.5)	385	15.7 (13.9, 17.5)	301	8.8 (7.7, 9.9)
Black	962	80.0 (77.3, 82.7)	98	8.7 (6.7, 10.7)	151	11.3 (9.1, 13.4)
Hispanic	1255	87.2 (84.9, 89.5)	93	7.0 (5.3, 8.8)	99	5.8 (4.3, 7.2)
Other or Missing	670	76.3 (72.9, 79.7)	105	12.6 (9.9, 15.2)	96	11.2 (8.2, 14.1)
Education							<.0001
Less than High School a	1541	75.4 (72.6, 78.3)	189	13.6 (11.0, 16.1)	227	11.0 (9.3, 12.7)
High School	1635	79.0 (76.8, 81.2)	230	13.0 (11.2, 14.8)	173	8.0 (6.6, 9.4)
College or more	945	74.7 (71.6, 77.8)	194	18.0 (15.3, 20.8)	99	7.3 (5.4, 9.2)
Missing	935	83.3 (80.7, 86.0)	68	6.2 (4.3, 8.2)	148	10.5 (8.5, 12.4)
Marital Status							0.0005
Married a	1625	75.7 (72.9, 78.5)	270	16.5 (13.9, 19.2)	170	7.8 (6.2, 9.3)
Other or Missing	3431	78.7 (77.0, 80.3)	411	11.7 (10.4, 13.1)	477	9.6 (8.5, 10.7)
Annual family income as % of Fed. Poverty Level							<.0001
<100% a	1702	81.4 (79.5, 83.4)	67	3.2 (2.2, 4.2)	349	15.4 (13.5, 17.2)
100% to <125%	454	79.7 (75.2, 84.2)	34	7.7 (4.3, 11.2)	63	12.6 (9.2, 15.9)
125% to <200%	1154	81.7 (79.2, 84.3)	129	10.2 (8.0, 12.4)	126	8.1 (6.4, 9.7)
200% to <400%	1308	74.6 (71.9, 77.3)	312	20.6 (18.0, 23.2)	91	4.8 (3.5, 6.1)
>=400%	438	69.5 (64.6, 74.4)	139	26.2 (21.4, 31.0)	18	4.3 (1.9, 6.6)
Employment							<.0001
Employed a	3194	77.1 (75.4, 78.8)	607	17.6 (16.0, 19.2)	236	5.3 (4.4, 6.2)
Not employed or Missing	1862	79.1 (76.9, 81.4)	74	4.7 (3.3, 6.0)	411	16.2 (14.3, 18.1)
Region							0.0002
Northeast a	454	71.9 (67.2, 76.5)	73	15.0 (10.9, 19.1)	110	13.1 (9.8, 16.5)
Midwest	1022	75.4 (72.7, 78.0)	172	13.2 (11.1, 15.3)	164	11.4 (9.4, 13.4)
South	2590	80.8 (78.9, 82.6)	307	12.5 (10.7, 14.2)	257	6.8 (5.6, 7.9)
West	990	77.6 (73.8, 81.3)	129	13.7 (10.2, 17.2)	116	8.7 (6.7, 10.7)
Health							<.0001
Good or Better Health a	4085	77.9 (76.4, 79.4)	606	14.5 (13.1, 15.9)	426	7.6 (6.7, 8.5)
Not “Good or Better” Health or Missing	971	77.3 (74.5, 80.0)	75	7.1 (5.4, 8.8)	221	15.6 (13.3, 17.9)
Flu Shot							<.0001
Received Flu Shot a	604	65.2 (61.4, 69.0)	136	18.3 (15.2, 21.4)	157	16.5 (13.7, 19.3)
Did not receive Flu Shot or Missing	4452	79.7 (78.2, 81.2)	545	12.4 (11.1, 13.7)	490	7.9 (7.0, 8.8)
Total Healthcare Spending in Last 12 Months b							<.0001
$0 a	2300	89.6 (88.0, 91.2)	167	7.7 (6.3, 9.1)	70	2.8 (1.9, 3.6)
$1–$500	1334	79.6 (77.1, 82.0)	176	12.6 (10.6, 14.7)	140	7.8 (6.3, 9.4)
$501–$1,000	370	68.7 (64.2, 73.2)	91	21.9 (17.3, 26.4)	75	9.4 (7.2, 11.7)
$1,001–$5,000	752	67.7 (64.0, 71.4)	176	19.7 (16.4, 22.9)	180	12.7 (10.6, 14.8)
$5,001–$15,000	225	56.0 (49.2, 62.8)	42	12.8 (8.3, 17.3)	131	31.2 (26.0, 36.5)
>$15,000	75	43.3 (33.1, 53.5)	29	25.0 (15.5, 34.6)	51	31.7 (21.1, 42.2)
Panel							<.0001
2003–2004 a	490	77.6 (73.4, 81.2)	72	15.2 (11.4, 19.0)	54	7.2 (5.1, 9.3)
2004–2005	459	79.5 (75.6, 83.4)	80	14.5 (11.1, 17.9)	47	6.0 (3.8, 8.2)
2005–2006	452	81.3 (76.9, 85.7)	53	12.2 (8.3, 16.0)	47	6.5 (4.2, 8.9)
2006–2007	475	80.7 (76.2, 85.1)	59	12.6 (8.4, 16.8)	52	6.8 (4.8, 8.8)
2007–2008	320	73.5 (68.1, 79.0)	54	17.0 (11.9, 22.1)	47	9.5 (6.6, 12.4)
2008–2009	544	79.7 (75.2, 84.2)	61	11.2 (7.4, 14.9)	64	9.1 (6.4, 11.9)
2009–2010	509	84.0 (80.3, 87.8)	46	8.3 (5.4, 11.1)	60	7.7 (5.2, 10.2)
2010–2011	426	80.1 (76.1, 84.1)	39	9.1 (6.1, 12.1)	50	10.8 (7.4, 14.2)
2011–2012	512	76.9 (72.8, 81.0)	76	12.7 (9.9, 15.6)	65	10.4 (7.0, 13.8)
2012–2013	516	80.6 (76.2, 85.0)	62	11.4 (7.6, 15.2)	65	8.0 (5.8, 10.2)
2013–2014	353	59.3 (53.4, 65.2)	79	22.3 (16.5, 28.1)	96	18.4 (14.1, 22.7)
Discussing Quitting Smoking with Provider							<.0001
Yes a	1585	73.3 (70.8, 75.8)	252	14.5 (12.4, 16.6)	306	12.2 (10.4, 14.0)
No	1962	78.7 (76.6, 80.8)	255	12.6 (10.7, 14.4)	233	8.8 (7.3, 10.2)
No visits in last 12 months	1509	82.1 (79.8, 84.3)	174	12.5 (10.4, 14.5)	108	5.5 (4.3, 6.7)

^aReference level used in regression analyses of Table 2, 3 and 4

^bTo avoid multicollinearity, this variable was not used in analyses that included the smoking cessation provider discussion variable.

All analyses, including standard errors and statistical tests, were performed according to the MEPS’ published weighting strategy to account for the complex survey sampling design. Statistical significance was assumed at p<.05, and analyses were conducted using domain analysis and the “proc survey” procedures of SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

Sensitivity Analyses

A sensitivity analysis was conducted to assess the validity of the insurance status definition. Specifically, it is possible for individuals designated as “uninsured” at both Round 2 and Round 4 interviews to have some insurance coverage between these two rounds. Moving back and forth between insurance statuses, or “churning”, is common among some populations. Low-income individuals whose incomes can frequently fluctuate may enroll and dis-enroll from the Medicaid program frequently.²⁸ The sensitivity analysis was performed by repeating all logistic regression analyses without individuals who were originally designated as “Remained Uninsured” but had some insurance coverage between the Year 1 and 2 interview periods.

A second sensitivity analysis was conducted to assess if collapsing the original three-level discussion with provider categorization to a two-level categorization would meaningfully change the results of the logistic regression analyses that included this variable. The survey question “In the past 12 months did a doctor advise you to quit smoking?” has three response levels: “Did have discussion with provider”, “Did not have discussion with provider”, and “Had no visits in the last 12 months”, which were used in the original analyses. The sensitivity analysis was performed by repeating logistic regression analyses which included the variable of smoking cessation discussion with a provider in the model with the two negative categories (“Did not have discussion with provider” and “Head no visits in the last 12 months”) collapsed into one category.

A third sensitivity analysis was performed to evaluate the impact of the exclusion of 562 individuals, who did not answer the question of discussion with a provider about smoking cessation, from the analytic sample. We repeated the logistic regression analysis with all 6,946 individuals (equal to 6,384 in the analytic sample plus 562) to study the impact of gaining private or public insurance on the likelihood of quitting smoking compared to remaining uninsured.

The first (i.e., check for validity of insurance status operationalization) and second (i.e., collapsing of the provider discussion variable) sensitivity analyses used all 6,384 individuals in the original analyses. The third sensitivity analysis (n=6,946) included an additional 562 individuals who were excluded from the original analyses because of a missing value for having a provider discussion about smoking.

Results

Of the 6,384 respondents in the analytic sample, 5,056 (77.8 weighted %) individuals remained uninsured, 681 (13.2 weighted %) individuals gained any private insurance, and 647 (9.0 weighted %) individuals gained only public insurance between Year 1 and Year 2 of their panel data.

There were demographic differences among smokers who remained uninsured and those who gained any public or private insurance (Table 1). A lower percent of females (72.7 weighted %) remained uninsured compared to males (81.0 weighted %), and a higher percent of females (12.8 weighted %) gained public insurance compared to male smokers (6.6 weighted %). Hispanic individuals were more likely to remain uninsured (87.2 weighted %) and less likely to gain public insurance (5.8 weighted %) than any other racial group. Furthermore, Hispanic respondents were less likely (7.0 weighted %) than White respondents (15.7 weighted %) or those with a Missing or Other race (12.6 weighted %) to gain private insurance. While the rates of remaining uninsured were about the same, individuals who reported good or better health were much more likely to gain private insurance than those who did not report good or better health and vice versa was true for gaining public insurance. Those who received a flu shot or were interviewed during the very last panel (2013–2014) were more likely to gain insurance than their counterparts.

Table 2 provides the results from the logistic regressions comparing smoking cessation between individuals who gained private or public insurance to those who remained uninsured. The adjusted logistic regression included all variables in Table 1 excluding “Discussion of Quitting Smoking with Provider”. In the unadjusted model, individuals who gained either type of insurance were not statistically more or less likely to stop smoking compared to individuals who did not gain insurance. After adjusting for covariates, individuals who gained private insurance were statically more likely to stop smoking than those who remained uninsured (OR: 1.346, p=0.033); however, those who gained public insurance did not have a statistically significant difference in smoking cessation compared to those who remained uninsured.

Table 2.

Odds ratio between insurance status and quitting smoking, among those uninsured current smokers of age 18 and older at Year 1, Medical Expenditure Panel Survey Longitudinal Data Files 2003–2004 through 2013–2014 (n=6,384)

	Did Quit N (Weighted %)	Did Not Quit N (Weighted %)	Unadj OR	95% CI	p-value	Adj* OR	95% CI	p-value
Did not gain insurance	903 (15.0)	4153 (85.0)	Ref	Ref		Ref	Ref	Ref
Gained insurance
Any Private	131 (17.4)	550 (82.6)	1.192	(0.921, 1.543)	0.182	1.346	(1.024, 1.769)	0.033
Only Public	112 (15.0)	535 (85.0)	1.003	(0.762, 1.320)	0.985	0.994	(0.730, 1.355)	0.972

^*Adjusted for gender, age category, race, education, annual family income as a percent of federal poverty level, marriage status, employment, region, health status, flu shoot received in the past year, total healthcare spending in last 12 months, and panel number.

Table 3 provides results from the logistic regressions analyzing the association between smoking cessation discussions with a provider and quitting smoking. Adjusted models included all covariates displayed in Table 1 excluding “Total Healthcare Spending” to avoid multicollinearity issues. In the unadjusted model, individuals who did not have a discussion about smoking cessation with a provider (OR: 1.331, p=0.006) or who did not have visits within the last 12 months (OR: 1.301, p=0.011) were more likely to stop smoking than those who had a discussion with their provider. However, neither of these odds ratios were significant in the adjusted model.

Table 3.

Odds ratio between discussion with provider and quitting smoking, among those uninsured current smokers of age 18 and older at Year 1, Medical Expenditure Panel Survey Longitudinal Data Files 2003–2004 through 2013–2014 (n=6,384)

	Did Quit N (Weighted %)	Did Not Quit N (Weighted %)	Unadj OR	95% CI	p-value	Adj* OR	95% CI	p-value
Did have discussion with provider	322 (13.0)	1821 (87.0)	Ref	Ref	Ref	Ref	Ref	Ref
Did not have discussion with provider	482 (16.6)	1968 (83.4)	1.331	(1.085, 1.633)	0.006	1.174	(0.946, 1.457)	0.146
Had no visits in the last 12 months	342 (16.3)	1449 (83.7)	1.301	(1.064, 1.592)	0.011	1.056	(0.841, 1.325)	0.638

^*Adjusted for gender, age category, race, education, annual family income as a percent of federal poverty level, marriage status, employment, region, health status, flu shoot received in the past year, and panel number.

Table 4 provides results of the logistic regression models that included both primary independent variables (i.e., insurance status and smoking cessation discussion with a provider). Models were run to include only the two primary independent variables (Model 1 in Table 4) as well as to include the two primary independent variables adjusting for all covariates displayed in Table 1 excluding “Total Healthcare Spending” (Model 2 in Table 4). After controlling for covariates (Model 2 in Table 4), individuals who gained private insurance were more likely to quit smoking than those who remained uninsured (OR: 1.330, p=0.036).

Table 4.

Odds ratio between insurance status/discussion with provider and quitting smoking, among those uninsured current smokers of age 18 and older at Year 1, Medical Expenditure Panel Survey Longitudinal Data Files 2003–2004 through 2013–2014 (n=6,384)

	Model 1* Adj OR	95% CI	p-value	Model 2* Adj OR	95% CI	p-value
Insurance status
Remained Uninsured	Ref	Ref	Ref	Ref	Ref	Ref
Gained Private	1.210	(0.934, 1.567)	0.149	1.330	(1.019, 1.737)	0.036
Gained Public	1.041	(0.789, 1.373)	0.778	1.002	(0.748, 1.343)	0.987
Provider discussion
Did have discussion with provider	Ref	Ref	Ref	Ref	Ref	Ref
Did not have discussion with provider	1.338	(1.089, 1.645)	0.006	1.179	(0.950, 1.463)	0.135
Had no visits in the last 12 months	1.310	(1.068, 1.608)	0.010	1.063	(0.847, 1.336)	0.597

^*Model 1 includes only insurance status and discussion with provider about smoking cessation as independent variables. Model 2 additionally adjusted for gender, age category, race, education, annual family income as a percent of federal poverty level, marriage status, employment, region, health status, flu shoot received in the past year, and panel number.

The analyses also revealed significant associations with demographic variables and the odds of smoking cessation. Age, race, and region were significantly associated with differences in the odds of quitting smoking in each of the adjusted models (results not shown). For the fully-adjusted model including both insurance status and smoking cessation discussion with a provider, individuals who were Black (OR: 2.129, p=<.0001), Hispanic (OR: 3.796, p=<.0001), or reported a missing or other race (OR: 1.770, p=0.002) were more likely to stop smoking when compared to White respondents. Additionally, individuals who were aged 40–54 (OR: 0.563, p=<.0001) or older than 55 (OR: 0.605, p=0.002) were less likely to stop smoking compared to the youngest age group (age 19 to 29 years). Individuals who lived in the Midwest were statistically less likely to quit smoking compared with individuals from the Northeast (OR: 0.645, p=0.020).

Sensitivity Analyses

A sensitivity analysis was conducted in order to assess the impact of individuals who were considered to be in the “Remained Uninsured” insurance while still having at least one month of insurance between the months of his or her Round 2 and Round 4 interviews. Out of the 5,056 individuals who were in the “Remained Uninsured” group in the analytic sample, 205 (4.05%) of individuals had at least one month of health insurance coverage between the two interview dates. Removing these respondents from the analytic sample resulted in 6,179 individuals in this sensitivity analysis. We repeated logistic regression analyses using n=6,179 and obtained essentially the same results as shown in Table 2 to 4, which used the full analytic sample (n=6,384).

A second sensitivity analysis was used to determine the association between having a discussion with a provider about smoking cessation and quitting smoking after collapsing the “Did not have discussion with provider” and the “Had no visits in the last 12 months” responses into one level. The results from this sensitivity analysis were essentially identical to those in the original analyses (Tables 3 and 4) (results available upon request).

The third sensitivity analysis was to examine the impact of the exclusion of 562 individuals, who did not answer the question of discussion with a provider about smoking cessation, from the analytic sample. We again obtained very similar results as shown in Table 2, which used the original analytic sample (n=6,384).

Discussion

This study aims to assess the relationship between gaining insurance, discussions with providers about smoking cessation, and quitting smoking among uninsured adult current smokers. This study adds to the literature by evaluating the impact of gaining different types of health insurance on smoking cessation. Additionally, this study considers whether the impact of gaining insurance is influenced by the individual having a discussion with his or her provider about smoking cessation in the past 12 months. Previous evaluations of the causal impact of gaining insurance on smoking behaviors are inconclusive.^17–19,29 To our knowledge, this is the first study to analyze the relationships between gaining insurance of different types, having a discussion with a provider about smoking cessation, and quitting smoking using nationally representative data.

The findings from this study suggest that there may be a meaningful impact of the type of insurance gained on smoking cessation. Specifically, individuals who gained private insurance were more likely to stop smoking compared to individuals who remained uninsured; however, individuals who gained public insurance were not more or less likely to stop smoking compared to those who remained uninsured. One potential explanation of the increased smoking cessation rates among individuals who gained private insurance could be the possibility that privately-insured individuals may have greater utilization of other smoking cessation resources, such as pharmaceutical therapies.³⁰ Additionally, those who gain private insurance may be more likely to have additionally become employed between the two time periods, which could provide access to additional smoking cessation resources.²⁶

The results from this study suggest that there were not statistically significant differences in adjusted smoking cessation rates among individuals who had or who did not have discussions with their providers about smoking cessation. This finding does not discredit the potential ability for provider counseling to aid in smoking cessation efforts. It is possible that a positive response to the MEPS question “In the past 12 months did a doctor advise you to quit smoking?” does not adequately capture a true provider effort to encourage a patient’s smoking cessation. Our findings align with a previous study that used one panel of MEPS data and found no impact of discussing smoking cessation with a provider on smoking cessation.³¹

Measuring the effect gaining insurance on all aspects of healthcare utilization and health behavior can provide healthcare systems with the information necessary to create programs that most effectively improve individual and population health. Gaining insurance can impact health both directly, such as through increased access to healthcare providers and medications, as well as indirectly, such as by altering an individual’s risk perceptions. It has been suggested that increases in investments in health, such as insurance, may lower time preference rates.³² Smokers have consistently been found to have higher rates of time preference compared to non-smokers.^33,34 Lower time preference rates have been suggested as being related to lower smoking rates and higher rates of smoking cessation.³⁵ The possibility of gaining insurance as a mechanism for lowering time preference could potentially indicate effects of access to health insurance coverage on health behaviors in addition to increased access to healthcare resources.

In 2010, the Patient Protection and Affordable Care Act (ACA) was passed with multiple initiatives aimed to improve health insurance coverage including the option for states to expand their Medicaid coverage to families with income levels at or below 138% of the federal poverty level.³⁶ It is well documented that low-income individuals are more likely to smoke than individuals with higher incomes.³⁷ This study provides valuable information about the relationship between gaining insurance and smoking cessation, which may be particularly useful for policies regarding the low-income population and Medicaid expansion. For example, nearly 40% of the enrollees in Arkansas’s Private Option Medicaid Expansion program are current smokers.³⁸ Future studies could evaluate health behavior changes between low-income individuals who gained private health insurance coverage under the Arkansas Medicaid expansion and those who gained traditional Medicaid health insurance.

This study confirms the findings of other studies that suggest that socioeconomic characteristics were associated with the type of insurance that individuals gained.²⁶ This information is critical in order to create health insurance initiatives and smoking cessation programs that can most effectively improve health outcomes. For example, this information could be used to create programs that target those populations that were more likely to remain uninsured. The current study suggests that certain populations, such as Hispanic smokers, may be less likely to gain insurance of any type compared to other racial groups. Furthermore, if discussions with providers about smoking cessation are unable to consistently result in smoking cessation, public health professionals should consider mechanisms of gaining private insurance that may result in improved smoking cessation rates.

This study has several limitations. First, smoking cessation may not be able to be evaluated in the relatively short time period after gaining health insurance that was used in this study. Smoking has been suggested as being slower to change than other health-related behaviors.³⁹ Second, the use of the MEPS provides some limitations. The MEPS is a patient-reported survey, which could potentially inaccurately represent an individual’s true smoking behavior. Third, the responses to the question regarding smoking cessation discussions with a provider within the last 12 months was taken in Year 1, which may explain why this variable did not result in smoking cessation by Year 2. The discussion with providers about smoking cessation survey question was only available for current smokers. This prevented our analyses from using Year 2’s answers of this question, since those who quit smoking would not have a response for the smoking cessation discussions with a provider question in Year 2. Fourth, individual’s response to the smoking behavior question could be influenced by recent efforts established by the ACA to control tobacco use, such as tobacco surcharges on health insurance premiums.⁴⁰ This could potentially differentially impact reported smoking behaviors among individuals with different insurance types. Lastly, the small number of individuals who gained insurance of different types prevented us from using more granular insurance categories, such as “Medicaid” and “Medicare”. Using more granular insurance categories may provide additional information about differences in smoking cessation rates among different populations.^18,19,29.

Conclusions

This study suggests that individuals who gain private insurance may be at a particular advantage for smoking cessation. Additionally, the findings indicate that gaining insurance may lead to increased smoking cessation rates through a method other than discussing smoking cessation with a provider. Future analysis should evaluate these relationships using a larger sample population, allowing a longer time period for smoking cessation, as well as considering the impact of other smoking cessation resources, such as access to smoking cessation medications, as they relate to gaining insurance of different types.