The Impact of Fluctuations in Pack-Year Smoking History in the Electronic Health Record on Lung Cancer Screening Practices

To the Editor:

Low-dose computed tomography (LDCT) reduces lung cancer deaths,^{1, 2} but implementation of screening programs has been challenging.³ The electronic health record (EHR) can facilitate appropriate patient selection. However, the reliable assessment of an individual’s smoking exposure is hindered by data that may be incomplete, inaccurate,⁴ or underestimate true exposure.⁵

In our EHR, EpicCare (Epic Systems), clinicians may enter the number of packs smoked per day and the total years smoked, which the EHR then multiplies to determine and display the pack-years (PY) of exposure. When either field is updated, the EHR revises the PY of exposure in real time by replacing the previously recorded value. For example, a patient who has decreased smoking from one pack per day to one-half pack per day and has smoked for 35 years would have the PY of exposure revised by the EHR from 35 to 17. We questioned whether downward revision of a patient’s documented PY of exposure would reduce the likelihood of LDCT referral for lung cancer screening by primary care physicians.

We reviewed all primary care encounters at a large academic community health-care system over 4 years, including encounters with unscreened patients who were 55 to 80 years old, had a ≥ 30 PY smoking history recorded at the time of the encounter, and had not quit smoking ≥ 15 years previously. Multiple variable logistic regression was used to determine the likelihood of placement of a LDCT order during that encounter given the EHR-calculated PY exposure while adjusting for potential confounders. All analyses were performed using R, version 3.4.0 (R Foundation for Statistical Computing).

The study included 53,407 encounters in 8,185 patients. Patients were less likely to have an LDCT ordered when seen in an encounter in which the EHR reported that smoking exposure was < 10 PY despite being previously documented at ≥ 30 PY (adjusted OR, 0.66; 95% CI, 0.46-0.93) (Table 1). A sensitivity analysis including only patients who received LDCT (thereby excluding those who may have not been referred due to patient or provider preferences) revealed similar results (adjusted OR, 0.59; 95% CI, 0.40-0.83). The range of PY fluctuations across the study population is found in Figure 1.

Table 1.

Comparison of Encounters in Which LDCT Was and Was Not Ordered^a

Characteristic	LDCT Ordered (n = 686) No. (% of subgroup)	LDCT Not Ordered (n = 52,721) No. (% of subgroup)	Logistic Regression
			Adjusted OR (95% CI)	P Value
Age at time of encounter, y
55-59	201 (29.3)	16,695 (31.7)	Reference
60-64	222 (32.4)	15,212 (28.9)	1.2 (1.0-1.4)	.09
65-69	158 (23.0)	10,487 (19.9)	1.3 (1.0-1.6)	.03b
70-74	84 (12.2)	6,879 (13.0)	0.99 (0.74-1.31)	.94
75-80	21 (3.1)	3,448 (6.5)	0.49 (0.30-0.77)	.003c
Sex of patient
Female	356 (51.9)	28,036 (53.2)
Male	330 (48.1)	24,685 (46.8)
Race/ethnicity of patient
White	483 (70.4)	31,244 (59.3)	Reference
Black	160 (23.3)	17,599 (33.4)	0.64 (0.53-0.77)	< .001d
Hispanic	25 (3.6)	2,579 (4.9)	0.76 (0.49-1.12)	.19
Other	18 (2.6)	1,299 (2.5)	0.83 (0.50-1.29)	.44
Preferred language of patient
English	663 (96.6)	50,292 (95.4)
Spanish	8 (1.2)	1,409 (2.7)
Other	15 (2.2)	1,020 (1.9)
Insurance
Commercial	133 (19.4)	7,403 (14.0)	Reference
Medicaid	205 (29.9)	15,707 (29.8)	0.85 (0.68-1.07)	.16
Medicare	324 (47.2)	27,043 (51.3)	0.68 (0.55-0.86)	< .001d
Other	24 (3.5)	2,568 (4.9)	0.56 (0.35-0.86)	.010b
Last documented exposure
≥ 30 pack-years	487 (71.0)	37,578 (71.2)	Reference
20-29 pack-years	91 (13.3)	4,737 (8.9)	1.6 (1.2-2.0)	< .001d
10-19 pack-years	73 (10.6)	6,388 (12.1)	0.90 (0.69-1.15)	.41
< 10 pack-years	35 (5.1)	4,018 (7.6)	0.67 (0.46-0.93)	.022b
Provider LDCT ordering volume
First quartile (lowest)	13 (1.9)	4,826 (9.2)	Reference
Second quartile	32 (4.7)	13,209 (25.1)	0.86 (0.46-1.69)	.64
Third quartile	117 (17.1)	11,998 (22.8)	3.6 (2.1-6.8)	< .001d
Fourth quartile (highest)	524 (76.4)	22,688 (43.0)	8.3 (5.0-15.3)	< .001d
Provider’s specialty
Internal medicine	232 (33.8)	20,866 (39.6)
Gerontology	234 (34.1)	6,430 (12.2)
Family medicine	172 (25.1)	16,528 (31.3)
Medicine/pediatrics	48 (7.0)	8,897 (16.9)

LDCT = low-dose CT.

^aThe columns on the right demonstrate the results of the multivariate logistic regression. Variables that were not selected in a forward stepwise regression were not included in the model and are therefore not accompanied by an OR.

^bP < .05.

^cP < .01.

^dP < .001.

Figure 1.

A violin plot demonstrating the discrepancy between last pack-year documentation for each encounter compared with the maximum previously documented for patients seen after the electronic health record documented pack-year history was dropped to < 30 pack-years. The shaded areas represent the probability distribution of the maximum documented pack-years for each group and filled circles correspond to outliers. Box plots were overlaid to demonstrate the median value (horizontal bar), interquartile range (box), and 95% CI (vertical lines). EHR = electronic health record.

In conclusion, primary care providers were less likely to refer eligible patients for screening LDCT during an encounter following a reduction in documented smoking exposure from ≥ 30 PY to < 10 PY. These missed opportunities may harm patients and should prompt provider caution and a thoughtful redesign of EHR smoking documentation.