A Comparison of Smoking History in the Electronic Health Record With Self-Report

Abstract

Introduction:

Knowing patients’ smoking history helps guide who may benefit from preventive services such as lung cancer screening. The accuracy of smoking history in electronic health records remains unclear.

Methods:

This was a secondary analysis of data collected from a portal-based lung cancer screening decision aid. Participants of an academically affiliated health system, aged 55–76 years, completed an online survey that collected a detailed smoking history including years of smoking, years since quitting, and smoking intensity. Eligibility for lung cancer screening was defined using the Centers for Medicare and Medicaid Services criteria. Data analysis was performed May–December 2018, and data collection occurred between November 2016 and February 2017.

Results:

A total of 336 participants completed the survey and were included in the analysis. Of 175 participants with self-reported smoking intensity, 72% had packs per day and 62% had pack-years recorded in the electronic health record. When present, smoking history in the electronic health records correlated well with self-reported years of smoking (r =0.78, p≤0.0001) and years since quitting (r =0.94, p≤0.0001). Self-reported smoking intensity, including pack-years (r =0.62, p<0.0001) and packs per day (r =0.65, p≤0.0001), was less correlated. Of those participants eligible for lung cancer screening by self-report, only 35% met criteria for screening by electronic health records data alone. Others were either incorrectly classified as ineligible (23%) or had incomplete data (41%).

Conclusions:

The electronic health records frequently misses critical elements of a smoking history, and when present, it often underestimates smoking intensity, which may impact who receives lung cancer screening.

INTRODUCTION

Tobacco is the leading cause of morbidity and mortality in the U.S., responsible for more than 400,000 yearly deaths.^1–3 A patient’s smoking status affects recommendations for cholesterol treatment, aspirin prophylaxis, immunizations, and screening for vascular disease and cancer.^4–8 Therefore, obtaining an accurate smoking history in the electronic health record (EHR) is essential for delivering high- quality care.

Medicare “Meaningful Use” criteria require health systems to document smoking status in the EHR,⁹ and EHRs contain structured fields to capture granular smoking history (e.g., current status, start year, quit year, and packs per day [PPD]). Unfortunately, this is difficult in a busy clinical practice. How well EHRs capture smoking history is unclear. Most studies have only examined whether EHRs accurately identify current or former smokers and have reported sensitivities ranging from 78% to 95%.^10–12 One study examined accuracy of pack-years but was limited to patients referred for lung cancer screening (LCS).¹³ None of these studies examined the accuracy of years since quitting, an important eligibility criterion for LCS. The EHR’s ability to sufficiently guide LCS screening remains unknown.

To investigate the accuracy of the smoking history captured in EHRs, data are analyzed from a pragmatic trial of an LCS decision aid that captured participants’ self-reported smoking history and EHR-recorded smoking history.¹⁴

METHODS

The study was done in an academic health system network of 70 community-based primary care clinics in North Carolina. Details of the trial and decision aid have been published.^14,15 Briefly, between November 2016 and February 2017, the EHR was queried weekly to identify individuals aged 55–77 years who were scheduled to see a network primary care provider within the next 4 weeks and had no prior history of lung cancer or other disease predicting short life expectancy. Those identified as never smokers in the EHR were excluded. The first 1,000 identified individuals were sent an invitation via the patient portal to visit a web-based LCS decision aid called mPATH-Lung. On visiting mPATH-Lung, participants answered 5 questions about their smoking history to determine their eligibility for LCS based on Centers for Medicare and Medicaid Services criteria.¹⁶

Of the 1,000 individuals who received an invitation, 404 visited mPATH-Lung and 349 participants entered self-reported smoking data. Of those, 13 participants were never smokers, leaving 336 participants with self-reported smoking data included in the analysis.

This study was approved by the Wake Forest Baptist Health IRB (IRB00036974).

Self-reported smoking history was defined as data collected from participants using mPATH-Lung. EHR-recorded smoking history was defined as data collected by a healthcare provider during an in-person encounter. As smoking history may be recorded at multiple encounters, the encounter date that was closest to the date the patient used mPATH-Lung was chosen. The following elements were examined: (1) smoking status (current, former, or never), (2) smoking years (years of smoking and years since quitting), and (3) smoking intensity (pack-years and PPD).

Individuals were defined as eligible for LCS using published Centers for Medicare and Medicaid Services criteria (age 55–77 years, current smoker or quit within the last 15 years, and smoked at least 30 pack-years).¹⁶ Patients with missing key data elements were considered indeterminant for LCS. Frequency tables were constructed to summarize eligibility for LCS using both self-reported and EHR-recorded data, and the sensitivity and specificity of the EHR data were examined using the self-reported data as the gold standard. When both self-reported and EHR-recorded data were available, Spearman correlation coefficients were assessed to determine the monotonic association between the 2 smoking metrics and Wilcoxon signed-rank tests were used to assess the difference between the measures. Extreme outliers (e.g., 10 PPD) were considered missing for this analysis. Smoking history was considered accurate under the following parameters: years of smoking (± 3 years), years since quitting (± 3 years), PPD (± 0.25 packs), and pack-years (± 5 years). Statistical analysis was performed using SAS, version 9.4.

RESULTS

Nonresponders were more likely to be male and were slightly younger but were otherwise similar to participants (Table 1). Participants reported a smoking mean of 30.5 (SD=14.0) years and almost 1 PPD (×=0.96, SD=0.56). Among the 264 former smokers in the sample, mean years since quitting was 19.8 (SD=13.3) years.

Table 1.

Demographic Characteristics of Responders and Nonresponders (n=336)

Participant characteristics	Nonresponders,a mean (SD) or n (%), n=651	Responders (current and former smokers),b mean (SD) or n (%), n=336
Age, years	63.8 (6.0)	64.7 (5.8)
Sex
Male	348 (53.5%)	141 (42.0%)
Female	303 (46.5%)	195 (58.0%)
Race
White or Caucasian	550 (84.5%)	284 (84.5%)
Black or African American	90 (13.8%)	46 (13.7%)
Asian	2 (0.3%)	2 (0.6%)
Other	9 (1.4%)	4 (1.2%)
Ethnicity
Hispanic or Latino	5 (0.8%)	4 (1.2%)
Not Hispanic or Latino	645 (99.1%)	332 (98.8%)
Refused to answer	1 (0.1%)	0 (0%)
Insurance status
Commercial	303 (46.5%)	141 (42.0%)
Medicare	322 (49.5%)	184 (54.8%)
Medicaid	13 (2.0%)	6 (1.7%)
Uninsured	13 (2%)	5 (1.5%)
Smoking status in EHR
Current	115 (17.7%)	70 (20.8%)
Former	536 (82.3%)	266 (79.2%)
Unknown	0 (0%)	0 (0%)
Packs per day present in EHR	385 (59.1%)	213 (63.3%)

^aNonresponder defined as someone who did not complete the smoking history items in the mPATH-Lung decision aid.

^bThirteen responders indicated they never smoked and are excluded from the table above and all analyses.

EHR, electronic health record; mPATH, mobile patient technology for health.

Table 2 compares participants’ self-reported smoking history with the EHR. Smoking status (current or former smoker) was accurately reported in the EHR for >90% of participants; however, measures of smoking years and smoking intensity were missing or inaccurate for more than half of participants. Data on pack-years were present and accurate only 20% of the time in the EHR. The EHR under-reported participants’ smoking years and smoking intensity (Table 3). Self-reported years since quitting had the highest correlation with the EHR (r=0.94, p<0.0001) (Appendix Figure 1, available online).

Table 2.

Smoking History Availability and Accuracy, Self-Reported Versus EHR-Recorded

Self-reported smoking history	Data available in EHR, n (%)	Data available and accurate in EHR,a n (%)
Current smoker, n=72	72 (100)	67 (93.0)
Former smoker, n=264	264 (100)	261 (98.9)
Years smoked (±3 years), n=336	210 (62.5)	71 (21.2)
Years since quitting (±3 years), n=336	200 (59.5)	149 (44.4)
Packs per day (±0.25 packs), n=175b	126 (72.0)	70 (40.0)
Pack-years (±years), n=175b	109 (62.3)	35 (20.0)

^aMissing in the EHR is counted as inaccurate.

^bmPATH-Lung only asks for smoking intensity from those who are current smokers or quit within the last 15 years.

EHR, electronic health record; mPATH, mobile patient technology for health.

Table 3.

Self-Reported Versus EHR-Recorded Smoking History Comparison in Participants With Both Self-Reported and EHR-Recorded Data Available

Variable	Na	EHR-recorded, median (IQR)	Self-reported, median (IQR)	Difference, p-valueb	Spearman correlation (p-value)	% under-reported by EHR
Years smoked	210	25.0 (24.0)	31.5 (22.0)	<0.001	0.79 (<0.001)	75.7
Years since quitting	200	19.4 (23.4)	20.0 (22.0)	0.088	0.94 (<0.001)	43.0
Pack-years	109	25.0 (32.0)	34.5 (26.5)	<0.001	0.62 (<0.001)	74.3
Packs per dayc	126	1.0 (0.5)	1.0 (0.5)	0.006	0.65 (<0.001)	38.1

Note: Boldface indicates statistical significance (p<0.05).

^aNumber of observations with EHR-recorded and self-reported data.

^bWilcoxon signed rank test.

^cThe median and IQR are the same between EHR-recorded and self-reported; however, the distribution of packs per day differs.

EHR, electronic health record.

Smoking history recorded in the EHR identified 35% of patients who met Medicare criteria for LCS. Among those eligible for screening, the EHR incorrectly classified 23% as being ineligible, and 41% were missing data needed to determine eligibility (8% missing years since quitting only, 23% missing pack-years only, and 10% missing both). A total of 227 patients had enough data in the EHR to determine LCS eligibility. The sensitivity and the specificity of EHR data were 60% and 97%, respectively.

DISCUSSION

The EHR accurately classifies more than 90% of patients as ever smokers or never smokers; however, detailed elements of smoking history (e.g., smoking years and smoking intensity) are often missing or underestimated. Overall, the EHR identifies only one third of patients who qualify for LCS.

Annual LCS is recommended for patients aged 55–80 years who have a 30–pack-year smoking history and either currently smoke or have quit smoking in the past 15 years.¹⁶ This study has significant implications, as clinicians may make faulty decisions if they rely on inaccurate EHR documentation of smoking.¹⁷ Additionally, health systems using EHR data to identify patients who may benefit from LCS may miss potentially eligible patients.

Social desirability bias and financial incentives may affect the accuracy of EHR-recorded smoking history. When asked by a healthcare provider, patients may underestimate their smoking or focus on current attempts to cut down. Tobacco users have up to 50% increases in premiums compared with nonusers, encouraging under-reporting.⁹ In one study, 7.7% of those who reported current or past smoking on a questionnaire subsequently reported never having smoked at a later date.¹⁸ Additionally, others have shown that patients’ EHR-calculated pack-years fluctuate over time as patients cut back on smoking.¹⁷

Limitations

This study has limitations. The study sample is limited to those who use the Internet in a single health system. However, another study found a similar sensitivity and specificity of the EHR for determining LCS eligibility, suggesting these estimates may be stable.¹⁹ Recall bias could affect results if time elapsed between when smoking history was captured in mPATH-Lung and in the EHR, although analyses limiting the sample to participants with self-reported data collected within 60 days of the EHR-recorded data yielded similar results.

CONCLUSIONS

Across all measures of smoking years and intensity, incomplete documentation of smoking history in the EHR was found. Continuous validation efforts are needed to understand the quality of smoking documentation. Health systems would benefit from studying and developing strategies to ensure accurate EHR smoking history, such as developing mechanisms that allow patients to update smoking history characteristics themselves.