Reasoning Language Model for Personalized Lung Cancer Screening

Chuang Niu; Ge Wang

REVIEW 3 major objections 5 minor 25 references

Reasoning Language Model for Personalized Lung Cancer Screening

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A chain-of-thought-trained LLM reaches a 1-year lung cancer risk AUC of 0.926, beating Lung-RADS and its own base model.

desk verdict A reasonable application of CoT distillation to lung cancer risk, but the headline reasoning-vs-plain comparison is confounded and Table 2 implies a possible test-set overlap, so the central claim doesn't hold as stated. read the letter →

arxiv 2509.06169 v1 pith:2QQX7CR5 submitted 2025-09-07 cs.LG cs.AI

Chuang Niu , Ge Wang This is my paper · ORCID

classification cs.LGcs.AI

keywords lungcancerscreeningreasoninglanguagemodelchain-of-thoughtriskassessmentLung-RADSdistillationreinforcementlearningNLST

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that a language model trained to reason aloud can turn lung cancer screening reports into personalized risk scores that are both more accurate and easier to audit than the standard Lung-RADS system. The evidence comes from the National Lung Screening Trial cohort, where the reasoning model reaches a 1-year AUC of 0.926 versus 0.8497 for Lung-RADS and 0.6105 for the same base model without reasoning. The claimed mechanism is chain-of-thought: the model breaks the patient's imaging findings and medical history into risk factors, scores each one, and combines them in an explicit data-driven equation. Because the reasoning is written out, the model's factor weights can be inspected during training and clinical review. The authors also report that reinforcement learning alone improves the base model but does not match distillation from a stronger teacher model.

What carries the argument

The chain-of-thought risk-assessment trace is the load-bearing mechanism. The student model is distilled from a large teacher that produces either plain answers or CoT answers, filtered by rejection sampling against ground truth, and then optionally refined with a GRPO/DAPO-style policy objective using rule-based score, format, and length rewards. At inference the model outputs reasoning tokens between special think tags, followed by the answer text from which a deterministic parser extracts the risk score. This mechanism is what the paper credits for both the accuracy gain and the monitorability, because the reasoning tokens expose the factor weights.

What would settle it

Retrain the thinking model on exactly the plain model's 55,213 samples with the teacher shown the label, keeping chain-of-thought responses; if the 1-year AUC falls from 0.926 toward 0.8396, the confound explains the result. Separately, check whether any patient IDs in the 36,335-sample Test set appear in the 197,329-sample RL-Aug set, since the two tables report identical sample and label counts.

Watch

Extended reading notes

Core claim

The paper's central claim is that explicit reasoning supervision makes a small LLM significantly better at lung cancer risk prediction than the clinical baseline and than the same LLM without reasoning. On held-out NLST patients, the distillation-plus-thinking model reaches a 1-year AUC of 0.926; the plain distilled variant reaches 0.8396, the unmodified base model 0.6105, and Lung-RADS 0.8497. Reasoning is not incidental: the model is trained to emit a think block that decomposes the input into factors such as nodule morphology, smoking exposure, age, and co-morbid lung disease, assigns sub-scores and weights, and returns a final risk score through a transparent equation. The paper interpre

Load-bearing premise

The central claim is only as strong as the assumption that the two distilled models differ solely in chain-of-thought supervision — the thinking model was trained on about 3.6 times more samples and the teacher did not see the ground-truth label — and the identical counts reported for the RL-Aug and Test sets leave open a possible patient-overlap leak.

Editorial extensions

If this is right

Reasoning models beat Lung-RADS on every horizon tested, with the largest gain at 1 year, so short-term screening triage is the most promising clinical niche.
The written reasoning trace gives clinicians and developers a place to look when a prediction is wrong, which pure end-to-end risk models do not offer.
Distillation from a strong teacher is currently the most effective way to produce these traces; RL alone lifts the base model but trails distillation in this experiment.
Deployment would require calibrating the continuous risk score and mapping it to actions such as interval imaging, PET-CT, or biopsy, with thresholds tuned locally.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If the reasoning benefit survives a matched-data replication, chain-of-thought supervision could serve as an audit trail for regulatory approval of AI screening tools.
The factor weights the model emits could be converted into a standalone clinical nomogram, potentially preserving much of the accuracy where running an LLM is impractical.
The same distill-reason-then-optimize recipe may transfer to other structured screening settings, such as breast or colorectal cancer, whose current rules have similar sensitivity-specificity trade-offs.
CoT monitorability is a testable claim: clinicians shown the reasoning traces should catch injected errors more often than clinicians shown only the final score.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Worth a look for the application, not for the headline numbers. The paper adapts CoT distillation and GRPO-style RL to predict lung cancer risk from NLST text-based clinical records and reports, and the idea of making predictions monitorable through the model's reasoning is sensible. The pipeline is systematic: template-based input construction, teacher distillation from Qwen3-235B, rejection sampling, RL with a rule-based reward, and multi-year AUC evaluation. The 1-year 0.926 AUC is a large jump over the base model and Lung-RADS, if it held up.

The load-bearing comparison does not hold up. Distill-Think (197k samples) is compared against Distill-Plain (55k samples) to claim reasoning helps, but the two sets differ by 3.6x in size. The teacher prompt also differs: plain mode reveals the ground-truth label, thinking mode does not. Either confound could explain the AUC gap; the paper does not isolate reasoning. That makes the central claim unsupported as written.

Table 2 is worse. RL-Aug and Test have identical sample counts, positive/negative splits, and patient counts (36,335 / 2,328). If that is real, the RL training environment overlaps the held-out test set, and all downstream test numbers are invalid. The paper says the test set is patient-level held out, so this is either a printing error or a leak. It needs a correction before anything else. Also, Distill-Think has exactly the same counts as the raw Dataset (197,329) despite the text saying rejection sampling filtered wrong predictions. Could be a reporting slip, but it is another internal inconsistency.

The 'data-driven system equation' is never defined; it is the model's own text output. That makes the monitorability claim illustrative, not validated. The paper does list sensible limitations (external validity, report synthesis, reward design, calibration), which is in its favor.

Overall: the engineering is real and the task is relevant, but the evidence for the headline claim is currently too shaky. It deserves a serious referee only after the dataset table is fixed and the Distill-Plain/Think comparison is rerun on matched data. I'd send it back for major revision, not desk-reject.

Referee Report

3 major / 5 minor

Summary. The paper proposes a reasoning language model (RLM) for personalized lung cancer risk assessment using NLST data. It constructs large text-format datasets from structured radiology findings and patient risk factors, distills teacher responses in both plain-answer and chain-of-thought modes, fine-tunes a 1.7B Qwen3 model, applies a GRPO-style reinforcement learning stage, and evaluates 1–6 year lung cancer risk prediction via AUC. The headline result is a 1-year AUC of 0.926 for the distilled reasoning model (LCS-1.7B-Distill-Think), compared to 0.8396 for the plain distilled model and 0.8497 for Lung-RADS. The authors argue that chain-of-thought reasoning improves both predictive accuracy and monitorability.

Significance. If the result held, the paper would show that a small language model can produce clinically competitive and auditable risk scores by integrating longitudinal imaging findings with individual risk factors, and that CoT reasoning specifically drives the improvement. The work assembles a large NLST-derived dataset, a reproducible template conversion pipeline, and a worked example of a decomposable risk score. These are useful assets. However, the current evidence for the central claim is compromised by a training-set confound between the plain and thinking distillation conditions and by an apparent overlap between the RL training set and the test set. The statistical and methodological issues must be resolved before the claimed benefit of reasoning can be assessed.

major comments (3)

[Table 2 / §3.1] As printed, Table 2 reports identical sample and patient counts for RL-Aug and Test (36,335 samples, 2,328 patients), with identical positive/negative splits. §3.1 states that Test is held out at the patient level, while RL-Aug is the RL training environment. If the two sets are distinct, the table is missing a separate Test row and must be corrected. If they are not distinct, the RL-trained model is evaluated on its training set, invalidating the RL results in Figure 2/Table 3 and casting doubt on the integrity of the entire evaluation. This must be resolved before any test-set conclusion is drawn.
[Table 2 / §3.1] The central reasoning-benefit claim compares LCS-1.7B-Distill (AUC 0.8396) with LCS-1.7B-Distill-Think (AUC 0.9260). These models are trained on datasets that are not matched: Distill-Plain has 55,213 samples and 47,220 negatives, while Distill-Think has 197,329 samples and 189,335 negatives. The teacher protocol also differs: plain mode provides the ground-truth label to the teacher and requests a calculation without saying the label is known; thinking mode withholds the label. The 3.6x sample-size difference, the different negative-set composition, and the label-visible vs label-hidden protocol are all confounds. The experiment therefore does not isolate CoT supervision, and the claim that reasoning itself improves risk prediction is unsupported.
[§3.2 / Table 3 / Figure 2] The text repeatedly describes improvements as 'significant', but no confidence intervals, standard errors, or significance tests are reported; the AUCs are point estimates from a single split. With a test set containing 1,323 positives, sampling variability is not negligible. In addition, the Lung-RADS baseline (0.8497) is not described: the mapping from ordinal Lung-RADS categories to a score and the method for computing its AUC are absent. These omissions prevent the reader from verifying the statistical claims and the headline comparison.

minor comments (5)

[Eq. (2)] The product indices are inconsistent: both products use t, and r_k is undefined. The first product should range over reasoning tokens and the second over answer tokens; K and T appear swapped.
[Abstract / §3.2 / Figure 3] The 'data-driven system equation' is never defined. The equation in Figure 3 is a free-form model output with arbitrary weights; no derivation or validation procedure is given. The monitorability claim therefore rests on an illustrative example rather than a tested mechanism.
[Table 2] The table is typeset so that row labels and counts are ambiguous; the reader cannot determine which row corresponds to Test from the table alone. This needs a clear formatting correction.
[Model naming] Model names are inconsistent: Figure 2 uses 'LCS-1.7B', 'LCS-1.7B-Distill', 'LCS-1.7B-Distill-Think', while Table 3 uses 'LCS-1.7B' and 'LCS-1.7B-Think'; the RL model appears only in Figure 2.
[Typos] Typos include 'benefitial' (Sec. 1), 'desinged' (Sec. 2.4), and 'noncalcied' (Sec. 2.2).

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: the core AUC is measured against external NLST ground-truth labels, and no reported prediction reduces to its inputs by construction. The main issues are confounded comparisons and an underdefined 'data-driven system equation,' which are validity concerns rather than circularity.

full rationale

The primary prediction target is a held-out Test set evaluated with AUC against pathology-based ground-truth labels from NLST follow-up, so the headline results are not equivalent to the training inputs. Reward thresholds t1=0.45 and t2=0.55 are hyperparameters tuned on training behavior to mitigate reward hacking, not fitted parameters renamed as predictions. The 'data-driven system equation' shown in Figure 3 is a chain-of-thought artifact generated by the model, not a fixed equation in the method; while the monitorability claim is under-supported, this is an evidentiary gap, not a reduction of a prediction to its inputs. The only self-citation ([19], M3FM) is background and is not load-bearing for the derivation. The main weaknesses - the Distill-Plain vs Distill-Think comparison differing in dataset size (55,213 vs 197,329) and teacher label visibility, and the identical sample counts for RL-Aug and Test in Table 2 - are potential confounds and possible data leakage, but they do not make any reported quantity equal to its inputs by construction. Thus no significant circularity is present.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on the validity of the NLST labels, the fidelity of text conversion, and the reliability of the teacher's CoT traces. The most concerning is the assumed correctness of the data splits, which the table contradicts.

free parameters (4)

Score reward thresholds t1, t2 = 0.45, 0.55
Set in Eq. 7 to mitigate reward hacking around 0.5; tuned empirically to training behavior.
Length penalty threshold l_max = 9000
Eq. 9, empirically set to limit reasoning length.
Reward weights alpha, beta = 1, 1
Eq. 10, simply set to 1.
GRPO group size G = 8
Set in Section 2.4.

assumptions (3)

domain assumption NLST pathology-confirmed lung cancer within n years is a correct ground-truth for lung cancer risk
Labeling in Section 2.2; excludes patients without pathology confirmation or with insufficient follow-up.
domain assumption Template-based conversion from structured NLST fields to free text preserves the information needed for risk prediction
Section 2.2; the authors acknowledge this may not capture real-world report variability.
domain assumption The teacher model Qwen3-235B-A22B produces reliable reasoning traces that can be distilled
Section 3.1; filtered by thresholding the teacher's score at 0.5, but this only guarantees directional correctness, not reasoning quality.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Reasoning Language Model for Personalized Lung Cancer Screening." pith.science (2026). https://pith.science/paper/2QQX7CR5

@misc{pith2026250906169,
  author       = {Pith},
  title        = {Pith review of: Reasoning Language Model for Personalized Lung Cancer Screening},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2QQX7CR5}},
  note         = {Machine review of arXiv:2509.06169}
}

read the original abstract

Accurate risk assessment in lung cancer screening is critical for enabling early cancer detection and minimizing unnecessary invasive procedures. The Lung CT Screening Reporting and Data System (Lung-RADS) has been widely used as the standard framework for patient management and follow-up. Nevertheless, Lung-RADS faces trade-offs between sensitivity and specificity, as it stratifies risk solely based on lung nodule characteristics without incorporating various risk factors. Here we propose a reasoning language model (RLM) to integrate radiology findings with longitudinal medical records for individualized lung cancer risk assessment. Through a systematic study including dataset construction and distillation, supervised fine-tuning, reinforcement learning, and comprehensive evaluation, our model makes significant improvements in risk prediction performance on datasets in the national lung screening trial. Notably, RLM can decompose the risk evaluation task into sub-components, analyze the contributions of diverse risk factors, and synthesize them into a final risk score computed using our data-driven system equation. Our approach improves both predictive accuracy and monitorability through the chain of thought reasoning process, thereby facilitating clinical translation into lung cancer screening.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 18 canonical work pages

[1]

https://www.lung.org/research/ 12 trends-in-lung-disease/lung-cancer-trends-brief

Association, A.L.: Lung cancer fact sheet. https://www.lung.org/research/ 12 trends-in-lung-disease/lung-cancer-trends-brief

work page
[2]

In: Healthcare, vol

Amicizia, D., Piazza, M.F., Marchini, F., Astengo, M., Grammatico, F., Battaglini, A., Schenone, I., Sticchi, C., Lavieri, R., Di Silverio, B.,et al.: Systematic review of lung cancer screening: advancements and strategies for implementation. In: Healthcare, vol. 11, p. 2085 (2023). MDPI

work page
[3]

New England Journal of Medicine365(5), 395–409 (2011)

Team, N.L.S.T.R.: Reduced lung-cancer mortality with low-dose computed tomographic screening. New England Journal of Medicine365(5), 395–409 (2011)

work page 2011
[4]

New England journal of medicine382(6), 503–513 (2020)

Koning, H.J., Aalst, C.M., Jong, P.A., Scholten, E.T., Nackaerts, K., Heuvelmans, M.A., Lammers, J.-W.J., Weenink, C., Yousaf-Khan, U., Horeweg, N.,et al.: Reduced lung-cancer mortality with volume ct screening in a randomized trial. New England journal of medicine382(6), 503–513 (2020)

work page 2020
[5]

European journal of cancer118, 142–148 (2019)

Pastorino, U., Sverzellati, N., Sestini, S., Silva, M., Sabia, F., Boeri, M., Can- tarutti, A., Sozzi, G., Corrao, G., Marchian` o, A.: Ten-year results of the multicentric italian lung detection trial demonstrate the safety and efficacy of biennial lung cancer screening. European journal of cancer118, 142–148 (2019)

work page 2019
[6]

Annals of internal medicine162(7), 485–491 (2015)

Pinsky, P.F., Gierada, D.S., Black, W., Munden, R., Nath, H., Aberle, D., Kazerooni, E.: Performance of lung-rads in the national lung screening trial: a retrospective assessment. Annals of internal medicine162(7), 485–491 (2015)

work page 2015
[7]

JNCI: Journal of the National Cancer Institute113(8), 1044–1052 (2021)

Fedewa, S.A., Kazerooni, E.A., Studts, J.L., Smith, R.A., Bandi, P., Sauer, A.G., Cotter, M., Sineshaw, H.M., Jemal, A., Silvestri, G.A.: State variation in low-dose computed tomography scanning for lung cancer screening in the united states. JNCI: Journal of the National Cancer Institute113(8), 1044–1052 (2021)

work page 2021
[8]

Jama325(10), 971–987 (2021)

Jonas, D.E., Reuland, D.S., Reddy, S.M., Nagle, M., Clark, S.D., Weber, R.P., Enyioha, C., Malo, T.L., Brenner, A.T., Armstrong, C.,et al.: Screening for lung cancer with low-dose computed tomography: updated evidence report and systematic review for the us preventive services task force. Jama325(10), 971–987 (2021)

work page 2021

Show all 25 references

[9]

Lung cancer and personalized medicine: current knowledge and therapies, 43–57 (2016)

Rivera, G.A., Wakelee, H.: Lung cancer in never smokers. Lung cancer and personalized medicine: current knowledge and therapies, 43–57 (2016)

2016
[10]

Annals of the American Thoracic Society19(5), 799–806 (2022)

Triplette, M., Wenger, D.S., Shahrir, S., Kross, E.K., Kava, C., Phipps, A., Hawes, S.E., Cole, A., Snidarich, M., Crothers, K.: Patient identification of lung cancer screening follow-up recommendations and the association with adherence. Annals of the American Thoracic Societ...

2022
[11]

Journal of Thoracic Oncology17(1), 38–55 (2022) 13

Lin, Y., Fu, M., Ding, R., Inoue, K., Jeon, C.Y., Hsu, W., Aberle, D.R., Prosper, A.E.: Patient adherence to lung ct screening reporting & data system– recommended screening intervals in the united states: A systematic review and meta-analysis. Journal of Thoracic Oncology17(1...

2022
[12]

JAMA Network Open4(7), 2116233–2116233 (2021)

N´ u˜ nez, E.R., Caverly, T.J., Zhang, S., Glickman, M.E., Qian, S.X., Boudreau, J.H., Slatore, C.G., Miller, D.R., Wiener, R.S.: Adherence to follow-up testing rec- ommendations in us veterans screened for lung cancer, 2015-2019. JAMA Network Open4(7), 2116233–2116233 (2021)

2015
[13]

Journal of the American College of Radiology14(11), 1403–1411 (2017)

Glover IV, M., Daye, D., Khalilzadeh, O., Pianykh, O., Rosenthal, D.I., Brink, J.A., Flores, E.J.: Socioeconomic and demographic predictors of missed opportu- nities to provide advanced imaging services. Journal of the American College of Radiology14(11), 1403–1411 (2017)

2017
[14]

Journal of Thoracic Oncology 14(5), 784–792 (2019)

Tseng, C.-H., Tsuang, B.-J., Chiang, C.-J., Ku, K.-C., Tseng, J.-S., Yang, T.-Y., Hsu, K.-H., Chen, K.-C., Yu, S.-L., Lee, W.-C.,et al.: The relationship between air pollution and lung cancer in nonsmokers in taiwan. Journal of Thoracic Oncology 14(5), 784–792 (2019)

2019
[15]

Radiology290(2), 278– 287 (2019)

Wang, G.X., Baggett, T.P., Pandharipande, P.V., Park, E.R., Percac-Lima, S., Shepard, J.-A.O., Fintelmann, F.J., Flores, E.J.: Barriers to lung cancer screening engagement from the patient and provider perspective. Radiology290(2), 278– 287 (2019)

2019
[16]

arXiv preprint arXiv:2108.07258 (2021)

Bommasani, R., Hudson, D.A., Adeli, E., Altman, R., Arora, S., Arx, S., Bern- stein, M.S., Bohg, J., Bosselut, A., Brunskill, E., et al.: On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258 (2021)

2021 arXiv
[17]

Nature medicine25(6), 954–961 (2019)

Ardila, D., Kiraly, A.P., Bharadwaj, S., Choi, B., Reicher, J.J., Peng, L., Tse, D., Etemadi, M., Ye, W., Corrado, G.,et al.: End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nature medicine25(6), 954–961 (2019)

2019
[18]

Journal of Clinical Oncology41(12), 2191–2200 (2023)

Mikhael, P.G., Wohlwend, J., Yala, A., Karstens, L., Xiang, J., Takigami, A.K., Bourgouin, P.P., Chan, P., Mrah, S., Amayri, W.,et al.: Sybil: A validated deep learning model to predict future lung cancer risk from a single low-dose chest computed tomography. Journal of Clinic...

2023
[19]

Nature Communications16(1), 1523 (2025)

Niu, C., Lyu, Q., Carothers, C.D., Kaviani, P., Tan, J., Yan, P., Kalra, M.K., Whitlow, C.T., Wang, G.: Medical multimodal multitask foundation model for lung cancer screening. Nature Communications16(1), 1523 (2025)

2025
[20]

Advances in neural information processing systems35, 24824–24837 (2022)

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D.,et al.: Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems35, 24824–24837 (2022)

2022
[21]

arXiv preprint arXiv:2507.11473 (2025)

Korbak, T., Balesni, M., Barnes, E., Bengio, Y., Benton, J., Bloom, J., Chen, M., Cooney, A., Dafoe, A., Dragan, A., et al.: Chain of thought monitorability: A new and fragile opportunity for ai safety. arXiv preprint arXiv:2507.11473 (2025)

2025 arXiv
[22]

arXiv preprint arXiv:2402.128751(2024)

Li, Z., Liu, H., Zhou, D., Ma, T.: Chain of thought empowers transformers to 14 solve inherently serial problems. arXiv preprint arXiv:2402.128751(2024)

2024 arXiv
[23]

arXiv preprint arXiv:2402.03300 (2024)

Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y., Wu, Y., et al.: Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300 (2024)

2024 arXiv
[24]

arXiv preprint arXiv:2503.14476 (2025)

Yu, Q., Zhang, Z., Zhu, R., Yuan, Y., Zuo, X., Yue, Y., Dai, W., Fan, T., Liu, G., Liu, L., et al.: Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476 (2025)

2025 arXiv
[25]

arXiv preprint arXiv:1707.06347 (2017) 15

Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.: Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 (2017) 15

2017 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

https://www.lung.org/research/ 12 trends-in-lung-disease/lung-cancer-trends-brief

Association, A.L.: Lung cancer fact sheet. https://www.lung.org/research/ 12 trends-in-lung-disease/lung-cancer-trends-brief

work page

[2] [2]

In: Healthcare, vol

Amicizia, D., Piazza, M.F., Marchini, F., Astengo, M., Grammatico, F., Battaglini, A., Schenone, I., Sticchi, C., Lavieri, R., Di Silverio, B.,et al.: Systematic review of lung cancer screening: advancements and strategies for implementation. In: Healthcare, vol. 11, p. 2085 (2023). MDPI

work page

[3] [3]

New England Journal of Medicine365(5), 395–409 (2011)

Team, N.L.S.T.R.: Reduced lung-cancer mortality with low-dose computed tomographic screening. New England Journal of Medicine365(5), 395–409 (2011)

work page 2011

[4] [4]

New England journal of medicine382(6), 503–513 (2020)

Koning, H.J., Aalst, C.M., Jong, P.A., Scholten, E.T., Nackaerts, K., Heuvelmans, M.A., Lammers, J.-W.J., Weenink, C., Yousaf-Khan, U., Horeweg, N.,et al.: Reduced lung-cancer mortality with volume ct screening in a randomized trial. New England journal of medicine382(6), 503–513 (2020)

work page 2020

[5] [5]

European journal of cancer118, 142–148 (2019)

Pastorino, U., Sverzellati, N., Sestini, S., Silva, M., Sabia, F., Boeri, M., Can- tarutti, A., Sozzi, G., Corrao, G., Marchian` o, A.: Ten-year results of the multicentric italian lung detection trial demonstrate the safety and efficacy of biennial lung cancer screening. European journal of cancer118, 142–148 (2019)

work page 2019

[6] [6]

Annals of internal medicine162(7), 485–491 (2015)

Pinsky, P.F., Gierada, D.S., Black, W., Munden, R., Nath, H., Aberle, D., Kazerooni, E.: Performance of lung-rads in the national lung screening trial: a retrospective assessment. Annals of internal medicine162(7), 485–491 (2015)

work page 2015

[7] [7]

JNCI: Journal of the National Cancer Institute113(8), 1044–1052 (2021)

Fedewa, S.A., Kazerooni, E.A., Studts, J.L., Smith, R.A., Bandi, P., Sauer, A.G., Cotter, M., Sineshaw, H.M., Jemal, A., Silvestri, G.A.: State variation in low-dose computed tomography scanning for lung cancer screening in the united states. JNCI: Journal of the National Cancer Institute113(8), 1044–1052 (2021)

work page 2021

[8] [8]

Jama325(10), 971–987 (2021)

Jonas, D.E., Reuland, D.S., Reddy, S.M., Nagle, M., Clark, S.D., Weber, R.P., Enyioha, C., Malo, T.L., Brenner, A.T., Armstrong, C.,et al.: Screening for lung cancer with low-dose computed tomography: updated evidence report and systematic review for the us preventive services task force. Jama325(10), 971–987 (2021)

work page 2021