Variation in Reporting of Incidental Findings on Initial Lung Cancer Screening and Associations with Clinician Assessment

Abstract

Purpose:

To quantify the distribution, frequency, and clinical significance of incidental findings (IFs) on initial lung cancer screening (LCS) and the association of report characteristics with subsequent assessment.

Methods:

Health records of patients undergoing initial LCS 2015–2018 at the Minneapolis VA Health Care System were retrospectively reviewed for: demographics, Lung-RADS coding, IFs, and subsequent clinical assessment. IFs were considered potentially significant (SIFs) if they were likely to require any follow-up. High-risk SIFs were potentially malignant. Primary outcome was the SIF being addressed. Outcomes were analyzed using a mixed-effects model.

Results:

Patients (n=901) were primarily male (94.1%) smokers (62.1%) aged mean 65.2 years. IFs were extremely common (93.9%) with an average of 2.6 IFs per scan (n=2,296). 786 IFs (34.2%) were deemed likely SIFs. 58/786 (7.4%) were high-risk. 222/786 (28.2%) were addressed by the clinician, of which 104/786 (13.2%) underwent testing. Reporting of SIFs varied between radiologists, with at least one SIF in the impression on 24%–78% of LDCTs with the S-modifier, used to indicate the presence of a SIF, applied to 0–51% of reports. In the mutually adjusted model, radiologist recommendation (AOR 4.67 95% CI 2.23,9.76), high-risk finding (AOR 4.35, 95% CI 1.81, 10.45), and reporting in the impression (AOR 2.58 95% CI 1.28, 5.18) were associated with increased odds of the SIF being addressed.

Conclusions:

Radiologists vary in reporting of IFs on lung cancer screening. Further standardization of reporting of SIFs may improve this process, with the simultaneous goals of generating appropriate testing when needed and minimizing low-value care.

Summary statement:

We found considerable variation in the reporting of significant incidental findings, with elements of radiology reporting strongly associated with clinicians addressing incidental findings in the health record.

Background:

Despite efficacy,¹ uptake of lung cancer screening (LCS) with low-dose computed tomography (LDCT) has been slow.² One potential concern is the burden associated with the high rate of incidental findings (IFs) that are identified.³ Many IFs are of little clinical relevance, but can lead to additional testing, clinician workload, patient anxiety, costs and harms. However, some IFs, such as non-pulmonary malignancies, are clinically significant and require further testing.⁴

Per the American College of Radiology (ACR)-Society for Thoracic Radiology Joint statement on LCS interpretation, a “review of the entire examination for other potentially significant findings should be performed and reported.”⁵ When using Lung-RADS, radiologists are instructed to apply the “S” modifier to signal the presence of a significant incidental finding (SIF).⁶ However⁷, the appropriate scenario to apply the S-modifier is not always clear.⁸ Some findings are particularly challenging to categorize given high prevalence among the LCS-eligible population. The LDCT report should serve to clearly differentiate concerning SIFs that require further evaluation from IFs of minimal clinical significance. The utility of the S-modifier for communication with clinicians is also largely unknown.

We sought to 1) quantify the frequency, distribution, workload, and clinical outcomes of IFs and SIFs reported on initial LCS exams and 2) examine the associations of patient and report characteristics with subsequent clinician assessment and testing. We hypothesized that characteristics of the radiology report would be strongly associated with subsequent documentation and testing, and that radiologists would vary in their reporting of IFs and SIFs.

Methods

We conducted a retrospective chart review of all patients initially screened for lung cancer from December 2015-April 2018 at the Minneapolis VA Health Care System. Eligibility was determined per the 2014 United States Preventive Services Task Force criteria.⁹ Patients were referred by primary care providers (PCPs) and underwent LDCT per ACR technical specifications. Nodule-related findings were managed by a coordinator, who notified PCPs of LDCT results and the presence of IFs via templated notes, with further IF follow-up per PCP discretion. Radiologists were required to use Lung-RADS and were trained using standardized materials. Structured templates were provided to assist with the standardization of reporting. IRB approval was obtained from the Minneapolis VA IRB (## 1594468–2). Chart reviews were completed by one of two study team members (MC, KR) using a web-based data form.

LDCT reports were reviewed for nodule and incidental findings. Size, density, and location of the most suspicious nodule, Lung-RADS score and cancer diagnosis were recorded. A concerning lung nodule was defined as a Lung-RADS Category 3 or 4. S-modifier, demographics, and reading radiologist were administratively identified. Fifteen radiologists as well as an outside teleradiology service (4 total LDCTs) interpreted 901 initial LDCTs. One LDCT was missing administrative Lung-RADS data and radiologist and was excluded.

Identification of Incidental Findings and Determination of Significance:

IFs were defined as any non-nodule finding reported anywhere in the report (“findings” or “impression” sections). All IFs were recorded unless explicitly stated to be normal variation. IFs were considered potentially significant (SIFs) if they would be reasonably expected to require any follow-up, including clinic visits, testing, referrals, counseling, or tracking. Guidelines for the management of incidental renal stones, renal cysts/masses, abnormalities of the aorta, coronary calcification, and endocrine nodules^10–14 were reviewed to determine recommendations for follow-up.¹⁵ IFs were not considered SIFs if the radiologist stated that the finding was benign. IFs were initially categorized as potentially significant (SIF) if they were: 1) recommended by the radiologist to undergo additional testing or 2) coded with a Lung-RADS S-modifier 3) recommended to have follow-up per any of the above referenced guidelines. If significance was unclear, such as for findings without a recognized guideline, findings were preliminary coded as significant. Emphysema and coronary artery calcification were considered SIFs only if stated to be severe,¹⁶ recommended by the radiologist to have follow-up testing, or were the only IF on an S-coded report.

Final determinations of significance were made by one of two pulmonary physician reviewers (6 and 25 years of experience) after the initial coding by study staff. For S-coded LDCT reports with multiple IFs, non-significant findings were re-coded to avoid over-categorizing significance. Prior literature supports that radiologists often recommend testing for benign findings,¹⁷ therefore the physician reviewer recategorized as non-significant common age-related (e.g., degenerative joint disease) or benign findings (e.g., “simple cyst” “granuloma” “benign adenoma”) that would not be recommended to undergo follow-up, regardless of S-modifier or Radiologist recommendation for testing. Other IFs were coded as SIFs if they would reasonably be expected to require some follow-up (e.g. cirrhosis, significant pleural abnormalities). SIFs were further considered “high risk” if they represented potential malignancies (non-lung nodules, cysts, lesions and masses not identified as benign).

Outcome Assessment:

The EHR was reviewed up to 30 days after the LDCT to determine any documentation related to the IFs. The clinician was considered to have addressed the SIF (‘Clinician addressed’) if any consults, imaging tests, lab tests, clinic visits, procedures, or medication orders were recommended, placed, or completed in relation to the SIF, if the clinician addressed the SIF in the medical record, or if testing was recommended but declined. Records were further reviewed up to 6 months after the LDCT to determine the results of testing and the outcome of the SIF.

Statistical methods:

Patient and report characteristics were descriptively summarized along with their IFs and SIFs, stratified by significance. We summarized the types of completed tests and their results by organ system.

To examine the associations of patient and report characteristics with subsequent documentation and testing (‘clinician addressed’), we pre-specified 7 variables comprising both patient and report characteristics hypothesized to influence future evaluation. We fit a mixed effects logistic regression model using the Laplace method with SAS 9.4 Proc Glimmix^18–20 with seven patient and report characteristics as the independent fixed effects, subsequent documentation and testing as the dichotomized dependent variable, and individual scan and radiologist as potential random effects, due to possible clustering of SIFs and their assessment status by scan and radiologist. The covariance structure was assumed to be diagonal (a different variance component for each random effect) and the Wald z test was used to assess the covariance parameter estimates.^21,22 ICC (Intraclass Correlation Coefficient) was used to assess the random effects²³. Statistical significance was defined as p < 0.05 without adjustment for the number of comparisons.

To better understand between-radiologist variation, we summarized reporting characteristics stratified by radiologist, after excluding those who interpreted fewer than 10 LDCTs due to lack of sufficient data and calculated a coefficient of variation. To ensure that excluding low-volume radiologists did not affect conclusions, we repeated the model including all radiologists.

Results:

901 patients with a mean age of 65.2 years underwent initial LDCT, of whom nearly all were male (94.1%) and white (88.6%), and most currently smoked cigarettes (62.1%). Most (632/901) had low-risk lung nodule findings. Ten patients (1.1%) were diagnosed with lung cancer (stages: 3 Ia, 2 Ib, 1 IIa, 3 IIb, 1 IV). Nearly all patients (93.9%) had at least one IF, with an average of 2.6 (+/− 1.35) IFs per report. More than half of patients (62.5%) had at least one SIF, with an average of 0.87 (+/− 0.85) SIFs per patient. The S-modifier was applied to 21.6% of LDCTs. (Table 1)

Table 1:

Patient Characteristics Among Veterans who underwent initial Low-Dose CT (LDCT) Lung Cancer Screening (n=901)

Characteristics	N (%)
Patient Characteristics
Age in years	65.24 ± 5.61
Male	848 (94.1)
Race
White	798 (88.6)
Black	34 (3.8)
Other Minority Race	20 (2.2)
Unknown/Declined	49 (5.4)
Current Smoker	558 (62.1)
Report Characteristics
Incidental Finding Reported	846 (93.9)
Number of Incidental Findings per Patient	2.55 ± 1.35 (0–7)
At least 1 SIF* present	563 (62.5)
High Risk SIF present	57 (6.3)
SIFs per patient	0.87 ± 0.85 (0–5)
Lung-RADS Category£
1	46 (5.1)
2	586 (65.0)
3	117 (13.0)
4	140 (15.5)
S-Modifier Applied to LDCT	196 (21.6%)

^*SIF: Significant Incidental Finding

^£12 scans with no Lung-RADS code

Mean and standard deviation presented for continuous variables

Of the 2,296 total IFs reported on the 901 scans, 786 (34.2%) were deemed potentially significant. Of those, 58 (7.4%) were categorized as high-risk, with 6.3% of patients having a high-risk SIF. The most common IFs were pulmonary (48.7%) and cardiovascular (27.5%). See Supplemental Table 1 for a full listing of IFs. The reporting characteristics of IFs and SIFs varied considerably. (Table 2) Of the 1,510 IFs deemed not significant, 195 (12.9%) were reported in the impression, while the remainder were in the ‘findings’ only. Among these non-significant findings, radiologists suggested follow-up testing 9.5% of the time, though completion of follow-up testing was rare (23/1510, 1.5%). (See Supplemental Table 2)

Table 2:

Distribution of incidental findings (IFs) by organ system and reporting characteristics of incidental findings deemed likely clinically significant (SIFs) among 901 initial low-dose computed tomography scans for lung cancer screening

	Incidental Findings n (%)	Potentially Significant Incidental Findings n (%)
	Total	Total	S-Modifier Applied	Reported in Impression	Radiologist Recommended Follow-up	Clinician Addressed SIF†	Additional Testing Completed
Total IFs Reported	2296	786 (34.2%)	293 (37.3%)	435 (55.3%)	409 (52.0%)	221 (28.1%)	103 (13.1%)
High Risk IFs	58 (2.5%)	58 (7.3%)	35 (60.3%)	44 (75.9%)	46 (79.3%)	34 (58.6%)	30 (51.7%)
Endocrine	66 (2.9%)	55 (7.0%)	29 (9.9%)	40 (9.2%)	38 (9.3%)	28 (12.7%)	24 (23.3%)
Thyroid Abnormalities	28	25	14	18	18	15	14
Adrenal Findings	38	30	15	22	20	13	10
Gastrointestinal	297 (12.9%)	188 (12.9%)	84 (28.7%)	116 (26.7%)	112 (27.4%)	48 (21.7%)	27 (26.2%)
Liver and Biliary	152	115	60	83	78	30	17
Pancreatic	13	12	3	5	4	5	1
Spleen	27	10	3	2	3	2	1
Diverticulosis	6	3	0	2	2	0	0
Hernias	85	36	10	12	13	3	1
Esophageal and Gastric	14	12	8	12	12	8	7
Renal	84 (3.7%)	32 (4.1%)	20 (6.8%)	23 (5.3%)	23 (5.6%)	12 (5.4%)	12 (11.7%)
Stones	20	16	7	11	8	2	2
Masses and Cysts	64	16	13	12	15	10	10
Musculoskeletal	29 (1.3%)	17 (2.2%)	9 (3.1%)	16 (3.7%)	14 (3.4%)	6 (2.7%)	2 (1.9%)
Fractures	4	4	4	4	4	4	0
Lytic/Blastic Lesions	4	4	2	4	4	2	2
Degenerative/Other changes	21	9	3	8	6	0	0
Vascular	632 (27.5%)	72 (9.2%)	25 (8.5%)	41 (9.4%)	44 (10.8%)	25 (11.3%)	16 (15.5%)
Aneurysms	50	50	12	33	37	21	12
Coronary Artery Calcification	541	22	13	8	7	4	4
Other Atherosclerosis	41	0	0	0	0	0	0
Pulmonary	1117 (48.7%)	403 (51.3%)	118 (40.3%)	186 (42.8%)	167 (40.8%)	97 (43.9%)	18 (17.5%)
Emphysema	471	31	12	23	20	8	4
Granulomatous Disease	80	0	0	0	0	0	0
Other Structural Findings	299	108	31	42	41	24	6
Infectious/Inflammatory findings	213	212	57	87	77	38	3
Abnormal Lymph Nodes	54	52	18	34	29	27	5
Other *	71 (3.1%)	19 (2.4%)	8 (2.7%)	13 (3.0%)	11 (2.7%)	5 (2.3%)	4 (3.9%)

^*Includes findings not fitting above categories (e.g. skin changes, gynecomastia)

^†Clinician ordered additional testing, patient declined testing, or clinician documented no need for additional evaluation

Of the 786 SIFs, 37.3% had the S-modifier applied to the LDCT, 55.3% were reported in the impression, and 52.0% had radiologist recommendations for follow-up. Most of the 435 SIFs reported in the impression had follow-up recommendations (370/435, 85.1%), but some had testing recommendations only in the body of the report (39/351, 11.1%). 14.7% had follow-up recommendations, but the S-modifier was not applied to the LDCT. High-risk SIFs (n=58) were more likely to have the S-modifier applied (60.3%) and follow-up recommendations made (79.3%). Relatively few SIFs (221/786, 28.1%) were addressed by the clinician in the medical record. Of those, 8 patients refused testing and 49.8% (110/221) had documentation that testing was unnecessary. Of the 786 total SIFs, 103 (13.1%) underwent at least one additional medical contact due to the SIF (Table 2). Among High-risk SIFs, 58.6% (n=34) were addressed by the clinician.

Among the 13.1% of SIFs that underwent further evaluation, GI and endocrine abnormalities resulted in the most testing. (Table 3) Of the ten biopsies, three resulted in a diagnosis of extrapulmonary cancer (renal, gastric, thyroid). A new non-lung cancer diagnosis was detected in 0.5% of the patients with a SIF and 0.3% of patients overall.

Table 3:

Testing and results of evaluation for significant incidental findings found on initial Low-Dose CT for lung cancer screening by organ system

	Endocrine		Pulmonary		Gastrointestinal		Renal		Musculoskeletal		Cardiovascular		Other		Total
	n=55	n	n=403	n	n=188	n	n=32	n	n=17	n	n=72	n	n=19	n	n=786
Evaluation and Testing	Imaging	22	Imaging	8	Imaging	21	Imaging	11			Imaging	12	Imaging	4	78 Imaging
	Biopsies	7			Biopsy	1	Biopsy	1	Biopsy	1					10 biopsies
	Consults	2	Consults	5	Consults	2	Consults	2			Consults	5			16 consults
	Lab	1	Bronchoscopy	1	Labs	5			Lab	1	Lab	1			8 labs
	Partial thyroidectomy	1	Pulmonary function testing	7			Procedure	1			Hospitalization	1			2 surgeries/procedures
Outcomes	Thyroid Cancer	1	Foreign Body	1	Gastric Cancer	1	Renal Cell Carcinoma	1	Benign Biopsy	1	Myocardial Infarction	1			3 cancer diagnoses
	Hyper-thyroidism	1	Acute infection (Levaquin)	1			Stone Removal	1							36 new chronic disease
			New Chronic Disease		New Chronic		New Chronic				New Chronic		New Chronic
			Diagnosis	20	Disease	6	Disease	1			Disease	6	Disease	3

Restricting to the 11 radiologists who interpreted at least 10 LDCTs (n= 886 LDCTs), radiologists varied greatly in their reporting of IFs. Radiologists reported at least one IF on 69–100% of LDCTs, with a range of means of 1.2–2.9 IFs per LDCT. Radiologists reported at least one SIF in the impression of 0–75% of LDCTs. The use of the S-modifier varied as well, with radiologists applying it to 0–51% of LDCTs. Among reported SIFs, further testing was recommended 34–69% of the time. (Table 4)

Table 4:

Variation in Reporting of Incidental Findings on initial Lung Cancer Screening by Interpreting Radiologist^*

Radiologist	Scans Interpreted (n)	At Least One IF in the Report n (%)	At Least One IF in the Impression n (%)	Mean Number of IFs Reported	S-Modifier Applied n (%)	At least One SIF In Impression n (%)	Recommended Follow-up of SIF†‡ n (%)
Radiologist 1	58	52 (90)	29 (50)	2.8	12 (21)	22 (38)	29 (50)
Radiologist 3	73	66 (90)	47 (64)	2.8	0 (0)	31 (43)	42 (51)
Radiologist 5	13	9 (69)	4 (31)	1.4	0 (0)	4 (31)	3 (43)
Radiologist 6	201	196 (98)	123 (61)	2.9	102 (51)	100 (50)	119 (55)
Radiologist 7	31	30 (97)	13 (42)	2.1	0 (0)	8 (26)	9 (50)
Radiologist 8	15	12 (80)	6 (40)	1.5	2 (13)	4 (27)	3 (50)
Radiologist 10	12	12 (100)	8 (67)	2.3	0 (0)	5 (42)	4 (67)
Radiologist 11	246	227 (92)	60 (24)	2.6	17 (7)	50 (20)	57 (34)
Radiologist 12	40	38 (95)	31 (78)	2.6	1 (3)	26 (65)	33 (59)
Radiologist 14	13	10 (77)	6 (46)	1.2	3 (23)	6 (46)	6 (60)
Radiologist 15	184	180 (98)	107 (58)	2.4	57 (31)	84 (46)	95 (69)
Overall	886	832 (94)	434 (49)	2.6	194 (22)	340 (38)	400 (52)

^*Restricted to radiologists who interpreted at least 10 LDCTs

^†Among total SIFs reported

^‡Abbreviations: IF=incidental finding, LDCT=low dose computed tomography, SIF=clinically significant incidental finding

To quantify the variation in IF reporting, the coefficient of variation (CV) was calculated for 4 key reporting variables: percent of scans with at least 1 IF, at least 1 IF in the impression, at least 1 SIF in the impression, and S-modifier used. Percent of scans with at least 1 IF was the only variable that did not appreciably vary between radiologists (CV 11.2%, less than a commonly accepted cutoff of 30% variation). The other three variables varied appreciably between radiologists in both the full and restricted sets (CVs ranged from 32.0% to 144.0%).

In the mixed effects model examining predictors of a SIF undergoing further assessment, restricted to the 11 radiologists, radiologist, as a random effect, was found not to account for the variability in the outcome (ICC=0.0114, Wald z=0.61, p=0.27) in the null model and caused the estimated G matrix to be not positive definite with the full model which indicates needing a simpler covariance structure. However, the individual’s initial scan nested within radiologist random effect was found to account for variability, even more so in a null model with only one random effect (scan) (ICC=0.1625, Wald z=1.72, p=0.04). The final model was missing 4 SIFs due to missing lung nodule data (n=765) and did not include radiologist random effect for the reasons above.

Radiologist, use of S-modifier, patient age, presence of a concerning lung nodule (Lung-RADS 3 or 4), and total number of IFs present in the report were not associated with the clinician addressing the SIF. The presence of a radiologist recommendation (AOR 4.67 95%CI 2.23–9.76), reporting in the impression (AOR 2.58, 95%CI 1.28–5.18), and high-risk finding (OR 4.35, 95%CI 1.81–10.45) were independently associated with increased odds of being addressed by a clinician, after controlling for all the variables in the model (Table 5). In a sensitivity analysis including low-volume radiologists, neither the model results nor the conclusions differed appreciably.

Table 5:

Results of mixed effects model predicting whether clinician addressed significant incidental findings reported on initial low-dose CT scan for lung cancer screening n=765^*†

Fixed Effects Variables	OR (95% confidence Interval)	p-value
Patient Age (per year)	1.00 (0.96, 1.05)	0.995
Lung Rads Score (3 or 4 vs 1 or 2)	0.80 (0.48, 1.36)	0.411
Number of IFs‡ on the report	1.01 (0.82, 1.23)	0.953
Radiologist recommended follow-up (yes vs no)	4.67 (2.23, 9.76)	<.0001
S-Modifier Applied (yes vs no)	1.36 (0.83, 2.24)	0.228
High risk finding (yes vs no)	4.35 (1.81, 10.45)	0.001
Reported in the impression (yes vs no)	2.58 (1.28, 5.18)	0.008

^*Cases excluded without complete reporting or read by radiologist interpreting <10 low-dose CTs

^†Individual radiologist was excluded from the final model as a random effect due to lack of significant variability in the outcome (ICC=0.014, Var = 0.038) and receipt of G matrix not positive definite warning with inclusion in full model. Model includes individual scans as a random effect (ICC=0.163, Var=0.638). Errors are assumed to distribute as logistic with mean of 0 and variance of $\raisebox{1ex}{${\pi }^2$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$

^‡IF: incidental finding, ICC: intraclass correlation coefficient.

Discussion:

In this cohort of patients undergoing initial LDCT for LCS, nearly all had IFs, most had SIFs, but relatively few had the S-modifier applied. We identified considerable variability in how SIFs are reported. The provision of recommendations, use of the S-modifier, and overall quantity of SIFs reported varied considerably between radiologists, which is consistent with prior literature²⁴. Some follow-up recommendations were provided only in the body of the report. This may be problematic, as many clinicians do not read the entirety of radiology reports²⁵.

Few SIFs underwent further evaluation or yielded diagnoses. This supports that test-related workload generated by SIFs is moderate. The “cognitive workload,” however, was higher. Many SIFs required PCP assessment that no testing was needed, though most SIFs were not further discussed in the medical record. This may be attributable to the fact that most of these would undergo “expectant” management or could be addressed in the context of past medical history. Time constraints are a noted barrier for clinicians referring patients to LCS. SIF management adds to the PCP time investment in this process, including burden of documentation even if no action is needed²⁶. This may be even more of an issue in the private sector where patients are more likely to utilize multiple health systems. Investment in records portability and sharing of images and reports to ensure that LDCTs are always compared to prior imaging is essential to minimize excess SIF reporting. Almost 10% of IFs deemed unlikely to be significant had recommendations for testing that were inconsistent with guidelines or clinical practice. This is similar to previous data from the NLST where radiologists often recommended further imaging for renal, hepatic, and thyroid findings deemed benign in appearance.^17,13,27 If completed, such testing is low-value and could dilute the net cost-benefit of LCS.²⁸

Our findings are similar to previous studies, which identified a similarly broad list of IF types, high overall prevalence of IFs, and variable use of the S-modifier.^29–31 In the VA lung cancer screening demonstration project, despite all participating centers receiving the same training and materials and caring for similar patient populations, reported prevalence of at least one SIF varied between sites from 20 to 63%, a difference unlikely to represent true variation.³² “Real world” studies of LCS^33,34 have almost uniformly found a higher prevalence of SIFs than trial data.^{35,6 31–33 36,37,29} As in our study, the prevalence of extrapulmonary malignancy was less than 1% and nearly always less than 0.5%. The higher prevalence of SIFs found in our data and other non-trial settings likely reflects the more heterogeneous population of radiologists interpreting LDCTs, the lack of clarity surrounding what constitutes a SIF, and a population with more comorbidities than trial participants. The high prevalence of IFs may be necessary to consider in the shared decision-making process.

Per the current ACR quick guide for reporting of incidental findings on LCS³⁸, which summarizes multiple ACR white papers on follow-up of incidental findings,³⁹ most IFs require no follow-up. In our mixed effects model, we confirmed a strong association between the way SIFs are reported and subsequent evaluation, particularly for radiologist recommendations. The S-modifier was found to not be significantly associated with clinician behavior in further assessing SIFs when controlling for everything else in the model, suggesting it does not improve communication beyond the contents of the narrative report. Currently, variability is present in all aspects of reporting. Previous studies have reported only fair inter-observer agreement (42% agreement, kappa 0.2) as to what constitutes a SIF and requires an S-modifier.³⁴ Though we did not formally assess inter-observer agreement, we found a large variation in SIF reporting consistent with poor agreement. The 2018 Chest guidelines recommend that programs adopt systematic methods for the characterization and assessment of IFs⁴⁰ in LCS. Several algorithms based on literature and expert consensus have been published⁴¹ to guide reporting and evaluation of IFs. However, all such guides do not agree with each other or with other subspecialty guidelines, which may lead to confusion. For example, some subspecialty societies recommend testing for findings (e.g., non-obstructing renal calculi¹⁰) generally considered non-significant in LCS-specific guides. Furthermore, the relevance of reporting mild emphysema or bronchial wall thickening in this population of uniformly older heavy smokers is unclear and lacks consensus⁴². Broad use of consensus reporting and testing algorithms could ease the burden on patients, clinicians, and radiologists.⁴³

Our study has limitations, including the single-center design and the predominantly white male Veteran population. Veterans have high rates of medical comorbidities, additional exposures, and health behaviors that might increase the prevalence of some SIFs. We would not expect this to impact the characteristics of the radiology report, but the single center may have introduced site/training bias in the length of the report and inclusion of a high number of IFs. The VA is a nationalized health system, so there is no financial incentive to complete additional testing and providers have access to remote records, which may have influenced testing rates. Though we reviewed the medical record for six months after the LDCT, some SIFs may not have completed all diagnostic testing during that period. Thus the final outcome and workload may not have been captured, such as for indolent cancers. The criteria we used to classify significance may have missed some SIFs. Alternatively, some IFs may have been classified as significant that were not. However, we feel our criteria reflect the practice patterns of most clinicians as well as published literature. Identification of IFs was based on the report. Some IFs may not have been recorded by the radiologist.

Our study has several strengths including the extensive chart review encompassing the full radiology report and medical record, the physician review to classify significance, and the inclusion of low-value testing for low-risk findings. Our study is one of the first to include documentation of clinical reasoning or patient refusal as a part of the process of SIF evaluation, as these processes cannot be captured using administrative data.

In conclusion, we found that IFs are extremely common and variably reported on LCS reports, with clear links between reporting characteristics and subsequent evaluation. We found little evidence that the S-modifier as currently deployed supports communication. Systematic improvements in report structure such as reporting only significant findings in the impression, ensuring that all SIFs are in the impression, not including testing recommendations in the body of the report, and adherence to testing recommendations as outlined by the ACR would be expected to improve clarity in reporting. Continued dissemination of clear, evidence-based recommendations regarding what to do—or not do—are necessary.

Abbreviation list:

LDCT

Low-dose CT

LCS

Lung Cancer Screening

SIFs

significant incidental findings

IFs

incidental findings

ACR

American College of Radiology

CV

Coefficient of Variation