Implementation of Lung Cancer Screening in Primary Care and Pulmonary Clinics

Abstract

Background

Although low-dose CT (LDCT) scan imaging lung cancer screening (LCS) can reduce lung cancer mortality, it remains underused. Shared decision-making (SDM) is recommended to assess the balance of benefits and harms for each patient.

Research Question

Do clinician-facing electronic health record (EHR) prompts and an EHR-integrated everyday SDM tool designed to support routine incorporation of SDM into primary care improve LDCT scan imaging ordering and completion?

Study Design and Methods

A preintervention and postintervention analysis was conducted in 30 primary care and four pulmonary clinics for visits with patients who met United States Preventive Services Task Force criteria for LCS. Propensity scores were used to adjust for covariates. Subgroup analyses were conducted based on the expected benefit from screening (high benefit vs intermediate benefit), pulmonologist involvement (ie, whether the patient was seen in a pulmonary clinic in addition to a primary care clinic), sex, and race and ethnicity.

Results

In the 12-month preintervention phase among 1,090 eligible patients, 77 patients (7.1%) had LDCT scan imaging orders and 48 patients (4.4%) completed screenings. In the 9-month intervention phase among 1,026 eligible patients, 280 patients (27.3%) had LDCT scan imaging orders and 182 patients (17.7%) completed screenings. Adjusted ORs were 4.9 (95% CI, 3.4-6.9; P < .001) and 4.7 (95% CI, 3.1-7.1; P < .001) for LDCT imaging ordering and completion, respectively. Subgroup analyses showed increases in ordering and completion for all patient subgroups. In the intervention phase, the SDM tool was used by 23 of 102 ordering providers (22.5%) and for 69 of 274 patients (25.2%) for whom LDCT scan imaging was ordered and who needed SDM at the time of ordering.

Interpretation

Clinician-facing EHR prompts and an EHR-integrated everyday SDM tool are promising approaches to improving LCS in the primary care setting. However, room for improvement remains. As such, further research is warranted.

Trial Registry

ClinicalTrials.gov; No.: NCT04498052; URL: www.clinicaltrials.gov

Take-home Points.

Study Question: Do clinician-facing electronic health record (EHR) prompts and an EHR-integrated everyday shared decision-making (SDM) tool designed for routine incorporation into primary care improve low-dose CT scan imaging lung cancer screening ordering and completion?

Results: Clinician-facing prompts and an EHR-integrated everyday SDM tool resulted in significant increases in low-dose CT scan imaging ordering and completion in the primary care setting.

Interpretation: Clinician-facing prompts and an EHR-integrated everyday SDM tool are promising approaches to improving lung cancer screening in the primary care setting.

Lung cancer is the leading cause of cancer death in the United States.¹ Low-dose CT (LDCT) scan imaging lung cancer screening (LCS) in patients with significant tobacco use can reduce lung cancer mortality.^2,3 The United States Preventive Services Task Force (USPSTF) recommended offering LDCT scan imaging for high-risk patients in 2013⁴ and recently expanded the inclusion criteria.⁵ The Centers for Medicare and Medicaid Services also recommended using shared decision-making (SDM) because of substantial potential downsides to screening and wide individual variation in patients’ lung cancer risk, life expectancy, and potential net benefit from screening.⁶ In the National Lung Screening Trial, the number of lung cancer deaths prevented by LDCT scan imaging varied dramatically between patients in the highest quintile of risk vs those in the lowest quintile of risk.⁷ Despite the evidence of effectiveness and the USPSTF recommendations, implementation of LCS into clinical practice has been slow in the United States, with 6.5% of eligible individuals screened in 2020.⁸ In Utah, the screening rate in 2020 was around 1.9%.⁸

A review that focused on barriers to LCS engagement identified multiple provider barriers, including unfamiliarity with eligibility criteria and insurance coverage, difficulty identifying eligible patients, need for guidance on the management of screening results, skepticism about the benefits of screening, and insufficient time or knowledge to conduct SDM.⁹ SDM may be especially challenging because of a requirement from the Centers for Medicare and Medicaid Services that a decision aid be used to inform the screening process.⁶ Indeed, a recent cross-sectional study found that SDM is rarely carried out for LCS and often is of poor quality.¹⁰

To address these challenges, some health systems have implemented a centralized approach to improve SDM for LCS. For example, some Veterans Administration (VA) health systems have used screening coordinators to assist primary care clinicians in identifying and offering LDCT scan imaging screening to eligible patients.¹¹ Although such centralized LCS models hold promise, particularly around population-based screening approaches, important barriers remain. Health systems may not be able to afford establishing and operating a centralized approach, such as supporting a dedicated LCS clinic. In addition, some patients may prefer to discuss screening with their primary care clinician, and not all patients will want to schedule additional visits specifically for LCS. Thus, centralized approaches may limit patients’ access to LCS.¹²

Widely adopted cancer screening (eg, breast, cervical, and colorectal cancer screening) generally has been conducted using a decentralized approach, in which primary care clinicians play a key role in recommending screening to eligible patients. To implement a decentralized approach to LCS successfully, the challenges described above must be addressed in primary care contexts where many clinicians may not be familiar with the nuances of screening guidelines and may not feel adequately supported in providing SDM.^9,13,14 Even providers who are familiar with the guidelines and have easy access to web-based decision aids that facilitate SDM face multiple other barriers including the need to identify screening-eligible patients and to enter multiple data points manually into a decision aid to obtain patient-specific guidance.¹⁵Also major time and attention barriers to decentralized screening exist. Primary care providers have an average of 10 to 15 min with each patient¹⁶ and cannot routinely devote more than 1 or 2 min to SDM for LCS.¹⁷ Moreover, patients eligible for LCS often have multiple comorbidities requiring attention during a clinic visit.¹⁸

A promising approach to improving decentralized LCS in primary care is to leverage the electronic health record (EHR) to identify screening-eligible patients automatically and to support personalized SDM using an EHR-integrated SDM tool whose inputs are prepopulated with EHR data.¹⁵ Moreover, such an SDM tool would be more suitable for use in time-constrained primary care settings if it supported a faster, compromise everyday SDM approach focused on feasibility, as opposed to a more detailed full SDM approach (e-Table 1).^19,20 The conventional full SDM approach emphasizes a detailed initial presentation of information that can take ≥ 5 min followed by an explicit discussion of patient values and often omits a clinician recommendation.^19,20 In contrast, everyday SDM is designed to be delivered much more quickly, wherein a highly tailored recommendation is provided as a part of a brief initial presentation and where patient values and preferences can be considered implicitly or explicitly as directed by the patient.^19,20

The objectives of this study were (1) to evaluate whether a decentralized, multifaceted intervention consisting of clinician-facing EHR prompts and an EHR-integrated everyday SDM tool was associated with improvements in the ordering and completion of LCS; (2) to conduct subgroup analyses for patients stratified by four factors (personalized benefit level, visit type, sex, and race and ethnicity); and (3) to estimate EHR-integrated SDM tool use in the intervention phase.

Study Design and Methods

This pragmatic clinical trial included a 12-month preintervention (ie, baseline) phase (August 24, 2019-August 23, 2020) and a 9-month intervention phase (August 24, 2020-May 23, 2021). The study was approved by the University of Utah Institutional Review Board (Identifier: 00125797) and was registered with ClinicalTrials.gov (Identifier: NCT04498052). A waiver of consent was approved. Data were extracted from the University of Utah Health (UHealth) Enterprise Data Warehouse on December 8, 2022.

The study was conducted in 30 primary care and four pulmonary clinics at 13 UHealth locations. UHealth uses a decentralized approach to LCS, wherein front-line clinicians (primarily primary care clinicians and pulmonologists) refer eligible and interested patients to obtain LCS. UHealth’s LCS program is accredited by the American College of Radiology, with the Huntsman Cancer Institute maintaining a registry of patients who have received LCS.

We conducted user-centered design and in-depth workflow assessments to guide the development and refinement of the intervention following the Evaluation in Life Cycle of Information Technology framework.^15,21 The resulting intervention consisted of (1) clinician-facing EHR prompts and (2) an EHR-integrated SDM tool. EHR prompts for “lung cancer screening discussion” (for patients in need of LCS SDM) or for “lung cancer screening” (for patients who had elected to undergo screening) were shown in the patient summary on the left side of the EHR (Fig 1, item 1) and in the EHR’s Health Maintenance module (Fig 1, item 2). For both preintervention and intervention phases, the USPSTF 2013 guideline criteria were used. Within the Health Maintenance module, hovering over the topic (Fig 1, item 3) provided a tooltip on LCS eligibility criteria and information on how to access the SDM tool (Fig 1, item 4). Further details of the clinician-facing EHR prompts (Fig 1) and the LDCT scan imaging ordering screen (Fig 2) are provided in e-Appendix 1. The intervention was implemented in coordination with the institution’s LCS program.

Figure 1.

Electronic health record Health Maintenance prompts for lung cancer screening (LCS) and shared decision-making (SDM). 1 = prompt in Care Gaps section of the patient summary visible to all clinical users reminding to discuss LCS; 2 = Health Maintenance tab; 3 = Health Maintenance prompt to discuss LCS; and 4 = hover-over message to use SDM tool (the tooltip display uses 2021 United States Preventive Services Task Force criteria; during the study, 2013 criteria were displayed).

Figure 2.

Clinician-facing prompt to conduct shared decision-making (SDM) before initiating a lung cancer screening regimen. 1 = dynamic prompt stating that SDM tool should be used to satisfy Centers for Medicare and Medicaid Services requirements; and 2 = static informational text informing of United States Preventive Services Task Force screening criteria and prompt to use SDM tool.

The SDM tool was designed originally in the VA as a non-EHR integrated web-based tool (https://screenlc.com).²² This tool originally was designed to support full SDM, and it worked well when used by full-time LCS coordinators in the context of dedicated LCS SDM sessions at the VA.²² However, when the SDM tool was studied in VA primary care and later was reviewed by primary care clinicians at UHealth in preparation for integration with the EHR and routine primary care workflows, it became clear that an everyday SDM approach needed to be supported because of clinician time constraints in this context. Thus, the tool was adapted so that only the elements needed for everyday SDM were kept on the main website page (Fig 3, item 1), with the content relevant to full SDM moved to supplemental tabs (e-Fig 1).

Figure 3.

Electronic health record (EHR)-integrated lung cancer screening shared decision-making app. Numbers 1 through 8 refer to time-saving features: 1 = streamlined interface; 2 = identification of high-benefit patients; 3 = note generation; 4 = discussion scripts; 5 = EHR integration; 6 = input autopopulation; 7 = one-click ordering; and 8 = workflow orchestration. Legend of pictograph (collapsed by default): red circles = lung cancer deaths; green circles = lives saved by screening; yellow circles = invasive procedures resulting from false-positive results; purple circles = major complications resulting from such procedures.

In addition to reorganizing the SDM tool’s contents to facilitate everyday SDM, the SDM tool was integrated with the EHR using a standards-based framework known as Substitutable Medical Applications Reusable Technologies on Fast Healthcare Interoperability Resources.^15,23 As described in Table 1, the tool offered multiple time-saving features designed to enable efficient and effective everyday SDM. The tool enabled users to generate Centers for Medicare and Medicaid Services-required documentation (Fig 3, item 3) for copying into the clinical note. Users also could order LDCT scan imaging (Fig 3, item 7) and document the patient’s decision (Fig 3, item 8); these actions in turn updated the patient’s status in the EHR’s Health Maintenance module and associated EHR prompts. The SDM tool also provided a short, suggested script for the conversation tailored to the patient’s risk profile (Fig 3, item 4; e-Table 2). These scripts were designed specifically to facilitate personalized everyday SDM conversations.

Table 1.

Eight Time-Saving Features of the EHR-Integrated SDM Tool

Time-Saving Feature	Feature Description
1.Streamlined interface	SDM tool presents all core information in one screen, with supplemental information in optional tabs and drill-downs
2.Identification of high-benefit patients	Patients for whom screening should be recommended, rather than simply offered, are identified clearly
3.Note generation	SDM tool generates personalized SDM documentation that meets Centers for Medicare and Medicaid Services billing criteria and can be pasted into the note
4.Discussion scripts	SDM tool provides sample 30-s scripts for discussing LCS that are tailored to patient risk levels
5.EHR integration	SDM tool can be launched within the EHR’s native user interface, facilitating access within usual clinical workflows
6.Input autopopulation	SDM tool autopopulates required data inputs from the EHR
7.One-click ordering	SDM tool enables one-click ordering of LDCT scan imaging screening in the EHR
8.Workflow orchestration	SDM tool can orchestrate larger clinical workflows in EHRs; for example, the SDM tool can write data into the EHR to satisfy prompts to conduct SDM or to trigger annual prompts to conduct LDCT scan imaging screening

Features 1-8 are illustrated in Figure 3. EHR = electronic health record; LCS = lung cancer screening; LDCT = low-dose CT; SDM = shared decision-making.

Prior work has described how to individualize screening benefit-level prediction, as well as the importance of doing so.^19,24, 25, 26 The SDM tool visually presented the spectrum of net benefit among the screening eligible population, with the goal of encouraging screening more strongly for those with more anticipated net benefit, based on a higher lung cancer risk and adequate life expectancy (Fig 3, item 2). The CHEST LCS guidelines recommend several methods for identifying patients who are likely to benefit most from screening, including by using the Life Years Gained From Screening-CT model¹⁸ and a threshold of at least 16.2 days of life expected to be gained from screening, as well as using the Bach models^27,28 and a threshold of at least 5.2% 10-year risk of lung cancer developing.²⁹ A microsimulation study found that for approximately 52.9% of screening-eligible patients at highest risk, the benefits of screening overcame even highly negative patient views about screening.³⁰ Based on this study, the CHEST guidelines also suggest that providers consider delivering a stronger recommendation for screening for patients at higher risk.²⁹ During the course of the study, the SDM tool supported the 2013 USPSTF guidelines⁴ and individualized predictions of net benefit based on the Bach risk model.^27,28 After the study’s completion, the SDM tool was updated to support the 2021 USPSTF guidelines⁵ and the Life Years Gained From Screening-CT model.¹⁸ The update to the Life Years Gained From Screening-CT model was made to incorporate additional risk factors not included in the Bach model, in particular race. The inclusion of race is important because it can address racial disparities by encouraging screening more strongly for high-benefit Black patients.

Statistical Methods

The target population for the data analysis was patients who completed at least one primary care office visit during the study period, met 2013 USPSTF criteria for LCS (55-80 years of age, ≥ 30-pack-year smoking history, current tobacco use or quit smoking in the last 15 years),⁴ had not undergone chest CT scan imaging (low dose or otherwise) in the past year, and had not declined screening in the past 3 years. Exclusion criteria for the analysis were history of lung cancer before the visit date, chest CT scan imaging carried out in the past year, or structured EHR data from the past 3 years indicating the patient decided against screening. The inclusion and exclusion criteria were computed for each office and telehealth visit in primary care and pulmonary clinics. Patients missing the smoking data required for assessing USPSTF LCS eligibility were not included. Among eligible patients needing LCS, most patients also needed SDM. We considered patients to need SDM if they did not have structured EHR data from the past 3 years indicating SDM had been conducted with the patient.

We conducted three substudies as described herein. All analyses were performed using R version 4.2.0 software (R Foundation for Statistical Computing). P values of < .05 were considered statistically significant. Patient characteristics described in this article include screening benefit level, care in pulmonary clinics, sex, race and ethnicity, age, tobacco use, comorbidities (e-Table 3), Charlson Comorbidity Index,³¹ COVID-19 hospitalizations,³² insurance status, most frequently visited primary care clinic, and other factors. Details of patient characteristics calculations and aggregation across visits are described in e-Appendix 1. We used the generalized linear mixed-effect models to account for correlations among visits of the same patients when comparing patient characteristics in preintervention and intervention phases.

LDCT Scan Imaging Ordering, Completion, and Follow-Through

For substudy 1, the hypothesis was that introduction of the intervention would result in increased LDCT scan imaging ordering, completion, and follow-through. Specifically, the LDCT scan imaging ordering rate was defined as the proportion of all eligible patients for whom LDCT scan imaging was ordered within 30 days of the visit date and was not cancelled on the same day. The LDCT scan imaging completion rate was defined as the proportion of all screening-eligible patients (denominator) for whom LDCT scan imaging was completed (numerator). The LDCT scan imaging follow-through rate was defined as the proportion of patients with an LDCT scan imaging order (denominator) who followed through and completed the LDCT scan imaging (numerator). In both cases, the LDCT scan imaging was required to have occurred within 120 days of ordering.

To evaluate whether the intervention was associated with improvements in LDCT scan imaging ordering, completion, and follow-through, we used logistic regression. We used mixed-effect models to account for correlations between visits of the same patient for ordering and completion.³³ We adjusted for all patient characteristics using the covariate balancing propensity scores approach.³⁴ We calculated and reported the marginal outcome rates for both the preintervention and intervention periods. Finally, we used the Wald test to calculate P values for the estimated ORs. Improvements in covariate balance after propensity scores adjustment are summarized in e-Figure 2.

Subgroup Analyses

Subgroup analyses were performed for expected benefit from screening, pulmonologist involvement, sex, and race and ethnicity. Our SDM tool was designed under the rationale that those with a higher estimated net benefit should receive stronger encouragement to undergo screening. Thus, our a priori hypothesis was that implementing the tool would lead to a higher rate of screening completion among high-benefit patients compared with eligible but intermediate-benefit patients.

For this analysis, screening was considered to be of high benefit if patients had an estimated ≥ 16.2 days of life gained from undergoing three rounds of LDCT scan imaging screening.^18,30 Otherwise, screening was considered to be of intermediate benefit. Patients with a life expectancy of < 5 years (calculated based on factors including age, smoking intensity and duration, and comorbidities¹⁸) also were considered to have intermediate benefit because they were unlikely to derive a survival benefit from LCS.³⁵

Subgroup analysis based on whether patients received pulmonary care also was identified as a relevant analysis a priori with the hypothesis that this intervention would have more impact for primary care providers, because pulmonologists already are focused on diseases of the lung and because primary care providers routinely use Health Maintenance EHR prompts at UHealth whereas pulmonologists do not. Sex and race and ethnicity analyses were performed to ensure fairness, and we did not expect to find any difference. We chose sex and race and ethnicity for fairness analysis because these factors can be associated with disparities in cancer screening36, 37, 38 and because Black patients have an elevated risk of dying of lung cancer.³⁹

We first summarized the observed outcome rates for each of the subgroups. Then, we proceeded with regression analysis for all subgroup variables other than pulmonologist involvement and race and ethnicity, which lacked sufficient variability for analysis. We used the same outcomes and regression approach as for substudy 1 and added interaction terms between subgroup variables and the intervention variable. To visualize trends in outcomes by screening benefit level, we plotted observed and estimated monthly visit-level outcome rates using Poisson regression.⁴⁰

EHR-Integrated SDM Tool Use

Substudy 3 was an exploratory aim and was conducted without a statistical hypothesis. SDM tool use was assessed through tool launch times recorded in the EHR audit log. We reported SDM tool adoption by clinics and individual clinicians.

Results

Patient and visit characteristics are summarized in Table 2. Across the two phases, 1,435 patients were enrolled in the trial. COVID-19 hospitalization and patient portal use were higher in the intervention phase.

Table 2.

Patient Characteristics in the Preintervention and Intervention Phases

Patient Characteristics	Before Intervention (August 2019-August 2020)	Intervention (August 2020-May 2021)	P Value
No. of patients	1,090	1,026	…
Average no. of visits per patient	3.1	2.3	…
Screening benefit level			.619
High	731 (67.1)	676 (65.9)	…
Intermediate	359 (32.9)	350 (34.1)	…
Pulmonologist involvement			.079
No pulmonary visits	1,000 (91.7)	962 (93.8)
At least 1 pulmonary visit	90 (8.3)	64 (6.2)
Female sex	458 (42.0)	441 (43.0)	.751
Race and ethnicity			.827
Non-Hispanic White	944 (86.6)	902 (87.9)
Non-Hispanic Black	17 (1.6)	17 (1.7)
Hispanic	68 (6.2)	57 (5.6)
Other	61 (5.6)	50 (4.9)
Age, y	65.2 (6.6)	65.3 (6.6)	.991
Tobacco use
No. of cigarettes/d	25.0 ± 10.6	24.6 ± 10.1	.022a
No. of years smoked	38.9 ± 9.5	39.2 ± 9.1	.43
Current or former tobacco use			.422
Current tobacco use	574 (52.7)	555 (54.1)
Former tobacco use, quit < 15 y ago	516 (47.3)	471 (45.9%)
Years since last quit for former tobacco use	7.3 ± 4.3	7.6 ± 4.3	.219
Comorbidities and conditions
Angina pectoris	95 (8.7)	87 (8.5)	.775
Cancer	265 (24.3)	254 (24.8)	.845
Chronic bronchitis	158 (14.5)	140 (13.6)	.492
Chronic kidney disease	217 (19.9)	210 (20.5)	.819
COPD or emphysema	495 (45.4)	475 (46.3)	.746
Coronary artery disease	274 (25.1)	271 (26.4)	.575
Diabetes	315 (28.9)	303 (29.5)	.868
Dependence on medical equipment	5 (0.5)	8 (0.8)	.37
Heart attack	135 (12.4)	128 (12.5)	.989
Hypertension	718 (65.9)	686 (66.9)	.783
Liver disease	190 (17.4)	174 (17.0)	.76
Other heart diseases	306 (28.1)	299 (29.1)	.698
Stroke	114 (10.5)	113 (11.0)	.791
CCI	3.1 ± 2.8	3.2 ± 2.8	.815
Other factors
BMI, kg/m2	29.0 ± 6.9	29.0 ± 7.0	.88
Parent(s) with lung cancer history	67 (6.1)	65 (6.3)	.828
Patient portal used in the preceding year	637 (58.4)	691 (67.3)	< .001a
COVID-19 hospitalizations in Utah	7.8 ± 8.5	43.9 ± 19.5	< .001a
Insurance			.176
Commercial	330 (30.3)	301 (29.3)
Government	711 (65.2)	694 (67.6)
Self-pay	49 (4.5)	31 (3.0)
Needed LCS SDM	1,085 (99.5)	1,010 (98.4)	.02a

Data are presented as No. (%) or mean ± SD, unless otherwise indicated. CCI = Charlson Comorbidity Index; LCS = lung cancer screening; SDM = shared decision-making.

^aP < .05.

LDCT Scan Imaging Ordering, Completion, and Follow-Through

LDCT scan imaging ordering and completion rates are summarized in Table 3. Rates adjusted for covariates are summarized in Table 3. LDCT scan imaging ordering and completion increased from 7.1% to 27.3% (P < .001) and from 4.4% to 17.7% (P < .001), respectively. Adjusted ORs were 4.9 (95% CI, 3.4-6.9; P < .001) and 4.7 (95% CI, 3.1-7.1; P < .001) for LDCT scan imaging ordering and completion, respectively. The LDCT scan imaging follow-through rate did not increase.

Table 3.

LDCT Scan Imaging Ordering and Completion Rates in the Preintervention and Intervention Phases

Outcome	Before Intervention (August 2019-August 2020)		Intervention (August 2020-May 2021)		OR (95% CI)	P Value
	No./Total No. (%)	Estimated Rate	No./Total No. (%)	Estimated Rate
LDCT scan imaging ordering	77/1,090 (7.1)	7.0 (5.0-9.8)	280/1,026 (27.3)	26.9 (21.2-33.4)	4.9 (3.4-6.9)	< .001a
LDCT scan imaging completion	48/1,090 (4.4)	5.0 (3.3-7.4)	182/1,026 (17.7)	19.8 (14.8-25.9)	4.7 (3.1-7.1)	< .001a
LDCT scan imaging follow-through	48/77 (62.3)	65.1 (51.8-76.4)	182/280 (65.0)	65.3 (58.1-71.9)	1.0 (0.5-1.9)	.98
LDCT scan imaging ordering by screening benefit level						.086b
High benefit	60/731 (8.2)	9.4 (6.6-13.4)	196/676 (29.0)	29.7 (22.7-37.8)	4.1 (2.7-6.1)	< .001a
Intermediate benefit	17/359 (4.7)	2.6 (1.1-5.9)	84/350 (24.0)	19.5 (11.5-31.1)	8.9 (4.3-18.7)	< .001a
LDCT scan imaging completion by screening benefit level						.32b
High benefit	36/731 (4.9)	6.8 (4.4-10.4)	129/676 (19.1)	23.3 (16.9-31.3)	4.2 (2.6-6.7)	< .001a
Intermediate benefit	12/359 (3.3)	1.7 (0.6-4.7)	53/350 (15.1)	11.6 (5.7-22.1)	7.5 (3.0-18.2)	< .001a
LDCT scan imaging ordering by pulmonologist involvement
No pulmonary visits	63/1,000 (6.3)	...	261/962 (27.1)	...	...	...
At least 1 pulmonary visit	14/90 (15.6)	...	19/64 (29.7)	...	...	...
LDCT scan imaging completion by pulmonologist involvement
No pulmonary visits	39/1,000 (3.9)	...	168/962 (17.5)	...	...	...
At least 1 pulmonary visit	9/90 (10.0)	...	14/64 (21.9)	...	...	...
LDCT scan imaging ordering by sex						.805b
Female	27/458 (5.9)	6.3 (3.5-11.1)	107/441 (24.3)	23.5 (15.5-34.0)	4.5 (2.5-8.3)	< .001a
Male	50/632 (7.9)	7.5 (4.9-11.2)	173/585 (29.6)	29.1 (21.7-37.7)	5.1 (3.3-7.8)	< .001a
LDCT scan imaging completion by sex						.908b
Female	18/458 (3.9)	5.0 (2.6-9.4)	74/441 (16.8)	19.9 (12.5-30.3)	4.8 (2.4-9.5)	< .001a
Male	30/632 (4.7)	5.1 (3.0-8.3)	108/585 (18.5)	19.7 (13.5-27.8)	4.6 (2.7-7.8)	< .001a
LDCT scan imaging ordering by race and ethnicity
Non-Hispanic White	65/944 (6.9)	...	240/902 (26.6)	...	...	...
Non-Hispanic Black	2/17 (11.8)	...	6/17 (35.3)	...	...	...
Hispanic	5/68 (7.4)	...	15/57 (26.3)	...	...	...
Other	5/61 (8.2)	...	19/50 (38.0)	...	...	...
LDCT scan imaging completion by race and ethnicity
Non-Hispanic White	43/944 (4.6)	...	159/902 (17.6)	...	...	...
Non-Hispanic Black	1/17 (5.9)	...	5/17 (29.4)	...	...	...
Hispanic	2/68 (2.9)	...	9/57 (15.8)	...	...	...
Other	2/61 (3.3)	...	9/50 (18.0)	...	...	...

Because of the small sample size, for LDCT scan imaging follow-through, we used logistic regression without the random effect of patient identifier. The denominator is the subset of patients for whom LDCT scan imaging was ordered. Only observed rates are shown for pulmonologist involvement and race and ethnicity data because of the sparsity of the variable, which made it unsuitable for the regression analysis. LDCT = low-dose CT.

^aP < .05.

^bFor interaction.

Subgroup Analyses

LDCT scan imaging ordering and completion rates for patient subgroups are summarized in Table 3. Regarding screening benefit level, the LDCT scan imaging ordering and completion rates were higher in the high-benefit group, but the interaction effect was not significant. Figure 4 shows monthly LDCT scan imaging ordering and completion rates for eligible visits stratified by screening benefit level.

Figure 4.

Graph showing low-dose CT scan imaging ordering and completion rates in high-benefit and intermediate-benefit patients in the preintervention and intervention phases. Monthly ordering and completion rates are shown as dots, and the regression lines show the trends before and after the intervention.

Regarding pulmonologist involvement, the screening ordering rate in patients not seen by a pulmonologist was lower in the preintervention phase (6.3% vs 15.6%) and about the same in the intervention phase (27.1% vs 29.7%) compared with patients seen by pulmonologists. Subgroup analyses by sex and race and ethnicity showed that improvements occurred in all patient groups.

EHR-Integrated SDM Tool Use

In the intervention study period, the EHR-integrated SDM tool was used at least once in 10 of the 13 locations, including 13 of 30 primary care clinics and two of four pulmonary clinics (Table 4). One hundred two providers placed LDCT scan imaging orders for 274 patients who needed SDM. The tool was used by 23 of 102 providers (22.5%) who ordered LDCT scan imaging. The SDM tool was used for 69 of 274 patients (25.2%). The tool was used for 51 of 187 high-benefit patients (27.3%) and 18 of 87 intermediate-benefit patients (20.7%). Tool use was not associated with improved LDCT scan imaging follow-through.

Table 4.

EHR-Integrated SDM Tool Use in the Postintervention Period for Patients for Whom LDCT Scan Imaging Was Ordered

Use Context	SDM Tool Used		SDM Tool Not Used
	No./Total No. (%)	LDCT Scan Imaging Follow-Through Rate	No./Total No. (%)	LDCT Scan Imaging Follow-Through Rate
Locations	10/13 (76.9)	...	3/13 (23.1)	...
Clinics	15/34 (44.1)	...	19/34 (55.9)	...
Primary care clinics	13/30 (43.3)	...	17/30 (56.7)	...
Pulmonary clinics	2/4 (50.0)	...	2/4 (50.0)	...
Individual clinicians	23/102 (22.5)	...	79/102 (77.5)	...
Used the app for at least 5 patients	4/102 (3.9)	...	79/102 (96.1)	...
Patients	69/274 (25.2)	41/69 (59.4)	205/274 (74.8)	135/205 (65.9)
High benefit	51/187 (27.3)	30/51 (58.8)	136/187 (72.7)	92/136 (67.6)
Intermediate benefit	18/87 (20.7)	11/18 (61.1)	69/87 (79.3)	43/69 (62.3)
No pulmonary visits	67/262 (25.6)	40/67 (59.7)	195/262 (74.4)	127/195 (65.1)
At least 1 pulmonary visit	2/12 (16.7)	1/2 (50.0)	10/12 (83.3)	8/10 (80.0)

Data are presented as No./Total No. (%). EHR = electronic health record; LDCT = low-dose CT; SDM = shared decision-making.

Discussion

A multifaceted intervention targeted at primary care clinicians was associated with increased LDCT scan imaging ordering and completion. Previous studies have identified the need for EHR-integrated tools incorporated into clinical worflows.²² To our knowledge, this is the first study to assess the impact of an EHR-integrated everyday SDM tool for LCS. The SDM tool, Decision Precision+, can be implemented within EHR systems using the Substitutable Medical Applications Reusable Technologies on Fast Healthcare Interoperability Resources interoperability standard and is available as a free tool through the EHR app store of Epic, a market leader for EHR systems.⁴¹

Our intervention did not specifically target the follow-through rate for LDCT scan imaging orders, which was around 65% for both preintervention and intervention periods. Such follow-through rates are consistent with those of prior studies.⁴² Although this suboptimal follow-through from LDCT scan imaging orders may have been caused partially by the COVID-19 pandemic, further efforts are needed to improve follow-through among those for whom screening is ordered. Moreover, further research is needed to understand how an SDM process involving individualized risk communication and guidance impacts LDCT scan imaging follow-through rates in the specific context of LCS.

The EHR-integrated SDM tool was used for approximately one-quarter of LDCT scan imaging orders requiring SDM. This finding is suboptimal, but still superior to previously reported SDM and SDM tool use rates. In one study, only 9% of patients who underwent LDCT scan imaging screening had a separate SDM visit within 3 months of screening.⁴³ In another study evaluating the quality of clinician discussions with patients on initiating LCS, no evidence was found that decision aids, patient education materials, or other SDM tools were used.¹⁷ Finally, in a study aiming to increase SDM by enabling the ordering of about 40 patient decision aids, clinicians ordered such decision aids in only 3% of visits.⁴⁴

Many barriers to SDM persist, including limited time, overworked physicians, insufficient provider training, and clinical information systems that do not provide sufficient support.^9,10,14,45 Although SDM has shown efficacy in randomized clinical trials, the conduct of SDM remains low outside of the context of trials where SDM is supported directly by the research team.^46,47 For LCS, unique challenges exist, including the relatively low prevalence of screening-eligible patients in primary care compared with other common conditions such as hyperlipidemia or diabetes, which can reduce awareness and the development of clinician self-efficacy; the added complexity of needing to consider patient-specific net benefit estimates; and the reality that older individuals with heavy smoking histories often have multiple comorbidities that compete for clinician attention. In acknowledging the lack of time as a major barrier, the SDM tool was designed specifically to streamline SDM for LCS as much as possible (Table 1).

In terms of subgroup analyses, all eligible patients showed significant improvements in both LDCT scan imaging ordering and completion regardless of estimated benefit level. Further research is needed to understand the impact of different approaches to supporting LCS on screening uptake among patients with different levels of expected net benefit from screening. In particular, further research is needed on how best to communicate with intermediate-benefit patients to ensure that their decisions are consistent with their values and preferences.

The subgroup analysis by pulmonologist involvement found that the observed screening rate in patients not seen by a pulmonologist was lower than for those seen by a pulmonologist before this multifaceted intervention, but quite similar during the intervention period. Finally, fairness analyses showed that the intervention was beneficial for female participants and people from racial and ethnic minority groups, including patients who are Hispanic or Black.

This study has several limitations. First, we used a preintervention and postintervention study design with no parallel control subjects. In particular, the COVID-19 pandemic started in the middle of the preintervention phase, introducing an important confounder that may have contributed to lower LCS, especially in the intervention phase. For example, the COVID-19 pandemic may have reduced the capacity of pulmonary providers to see patients and to address screening needs. In addition, the lack of a concurrent control group introduces the possibility of a Hawthorne effect, with improvements occurring because clinicians are aware that they are being monitored. However, LCS was not an institutional clinical performance measure during the study, and the rollout of the intervention was communicated using typical channels for notifying clinicians regarding EHR system updates. Therefore, we do not believe the Hawthorne effect played an important role. Although we explored the use of a concurrent control group, this was not possible because some intervention components could not be patient-randomized or clinic-randomized; specifically, the EHR’s Health Maintenance features (Fig 1) must be turned on for all patients and all clinic locations, with no opportunity for selective activation. To address the limitations of the preintervention and postintervention study design, we used an advanced modeling technique (propensity scores) to adjust for changes in patient population. Although the propensity scores improved the balance for COVID-19 hospitalizations, residual imbalance remained that may have diluted the estimated effect of the intervention. Also, we visualized the trends to show that no upward trend in screening ordering and completion was present before the intervention.

Second, the study was performed in one academic health care system; as such, our findings may not be generalizable to other health systems. However, the intervention was implemented over Epic, a leader in the US market for ambulatory EHRs, and the SDM tool integrates with the EHR using the Substitutable Medical Applications Reusable Technologies on Fast Healthcare Interoperability Resources standard, which is supported widely across EHR vendors. These factors may facilitate scalable dissemination and further study of this intervention across many more health systems in the future.⁴⁸ Third, most of the patients in the study population were White. To mitigate this limitation, we performed a fairness analysis and found improvements in LDCT scan imaging ordering and completion rates for Hispanic and Black patients. Fourth, this study relied on smoking history data retrieved from the EHR database. A recent study by our group showed that using the last recorded EHR observation may underestimate patient eligibility for screening by up to 49% because of inaccurate and missing data.⁴⁹ Because this study did not include interventions specifically designed to improve smoking history documentation in the EHR, further research is needed on how to improve the quality of these underlying EHR data. However, we used similar data before and during the intervention phase. Fifth, we did not measure the quality of SDM. Finally, the SDM tool itself may have had limitations. Despite multiple attempts to prompt for the use of the SDM tool in the EHR, clinicians bypassed those prompts in many cases, indicating that more effective approaches may be needed for directing providers to appropriate use of the tool. At the same time, more stringent approaches to requiring use of an SDM tool (eg, not allowing placement of a LDCT scan imaging order unless SDM tool use has been documented), if not implemented optimally, inadvertently could result in appropriate patients not being screened because of the added burden. Although use of SDM tools can increase over time as clinicians become more aware of the tool,⁵⁰ more research is needed for how to increase the use of SDM in busy clinical settings.¹⁹ For example, we are exploring additional strategies for facilitating the conduct of personalized SDM for LCS, including using tools integrated with the EHR’s patient portal to engage patients directly.

Interpretation

Implementation of a multifaceted intervention including clinician-facing EHR prompts and an EHR-integrated SDM tool to personalize screening was associated with a fivefold increase in the odds of LDCT scan imaging ordering for eligible patients. Using a combination of an EHR-integrated everyday SDM tool and clinician-facing EHR prompts can allow successful implementation of LCS in a decentralized manner in primary care.

Funding/Support

The work reported in this article was supported in part by the Agency for Healthcare Research and Quality [Grants R18HS026198 and R18HS028791]. T. J. C. was supported by a VA HSR&D Career Development Award [Grant CDA 16-151].

Financial/Nonfinancial Disclosures

The authors have reported to CHEST the following: K. K. reports honoraria, consulting, sponsored research, writing assistance, licensing, or codevelopment in the past 3 years with Hitachi, Pfizer, RTI International, the University of California, San Francisco, Indiana University, the Korean Society of Medical Informatics, the University of Nebraska, NORC at the University of Chicago, the University of Pennsylvania, MD Aware, and the United States Office of the National Coordinator for Health IT (via Security Risk Solutions) in the area of health information technology outside of the submitted work; being an unpaid board member of the nonprofit Health Level Seven International health IT standard development organization, an unpaid member of the United States Health Information Technology Advisory Committee, and helping to develop a number of health IT tools that may be commercialized to enable wider impact. None declared (P. V. K., H. L., T. J. C., G. D. F., A. F., J. M. B., R. H., Y. Z., T. T., M. C. F., C. R., D. K. M., I. A. W., S. R.-L., P. B. W.).