AI Triage Expanded Lung Nodule Clinic Volume Across Three US Systems
A commercially available AI support tool changed the pace of lung nodule care in three US clinics. After implementation, the systems saw more new patients, more invasive procedures, and a higher monthly count of diagnosed cancers, even though the median time from baseline CT to diagnosis did not improve.
The study used a real-world before-and-after design across matched 12.5-month periods, comparing clinic activity before and after deployment of the software. The setting matters: pulmonary nodules are common incidental findings on chest CT, but only a subset represent early lung cancer, and the window for curative treatment closes quickly when triage is slow or inconsistent. In that context, any tool that can move more patients into specialty review has a direct operational consequence.
Across the three sites, new patient volume rose from 138 before implementation to 232 afterward. On a monthly basis, that translated to 11.0 patients before versus 18.6 after implementation, with the median rising from 9 to 16 patients. The difference was statistically significant. It also signals a tangible shift in clinic flow rather than a marginal change at the edges of the schedule.
Procedural Throughput Rose While Diagnostic Timing Held Steady
The increase was not limited to intake. Monthly patients undergoing at least one invasive procedure increased from 4.9 to 8.0 after the software went live. In practical terms, more nodules moved beyond surveillance and into tissue acquisition or other invasive evaluation pathways. For a lung nodule clinic, that usually reflects more decisive risk stratification and faster movement toward answers for patients who cannot wait on serial imaging alone.
A single sentence captures the operational tradeoff: the clinics processed more patients, but they did not diagnose them faster.
Median time from baseline CT to diagnosis was 58 days before implementation and 57 days afterward, a difference that was not statistically significant. The interquartile range narrowed from 125 days to 58 days, however, suggesting less spread in diagnostic timing after implementation. That pattern points to a more consistent pathway, even if the central tendency stayed flat. The software may have improved access and throughput more than it compressed the overall timeline from nodule detection to final diagnosis.
Cancer Detection Increased Without a Clear Statistical Signal
Monthly cancers increased from 3.0 to 3.8 after implementation, but this change did not reach statistical significance. Monthly early-stage cancers rose from 1.4 to 2.4, also without statistical significance. The direction of travel is clinically interesting because stage I and II disease is where intervention can still be curative, and a triage system that helps surface more of these patients could influence long-term outcomes if the signal holds in larger datasets.
Still, the numbers require restraint. The study did not show a confirmed increase in cancer detection or early-stage diagnosis at conventional thresholds, so the clearest evidence lies in clinic activity and procedure volume, not in a proven stage-shift effect. The biological and care-delivery logic is straightforward, though: if AI support helps identify higher-risk nodules more reliably and routes them into specialty care sooner, then more patients should reach biopsy, bronchoscopy, or surgical assessment rather than drifting in surveillance queues.
What the Before-and-After Design Can and Cannot Prove
This was a real-world evaluation, not a randomized trial, so causation cannot be claimed with the same confidence as in a controlled experiment. Changes in referral patterns, site workflow, staffing, local imaging volume, or institutional practice could also influence throughput. Even so, the matched pre- and post-implementation periods provide a useful look at what happened once the software entered routine care.
The findings are also shaped by the clinic setting. These were lung nodule clinics, not general pulmonary practices, which means the intervention acted on an already selected population with a pre-existing pathway for specialty review. In that environment, an AI tool does not need to diagnose cancer on its own to matter. It only needs to improve identification, triage, and downstream routing well enough to increase the number of patients who actually make it into the system and receive invasive evaluation when appropriate.
Commercial involvement sits in the background of the study and should be read carefully. The software was developed by a company with ties to the analysis, design, and manuscript preparation, while the study itself was funded through separate commercial research agreements. Those details do not invalidate the findings, but they do make independent replication especially important before any broad claims about patient-level benefit are made.
Why the Signal Still Matters for Lung Cancer Pathways
Lung cancer remains the leading cause of cancer mortality worldwide, and late diagnosis is a major reason why outcomes remain poor. Pulmonary nodules are often discovered incidentally, which creates a workflow problem: clinicians must decide which lesions can safely wait and which need immediate workup. Risk calculators such as Mayo and Brock help, but real-world practice still depends on human judgment, fragmented follow-up, and the ability of a clinic to absorb referrals without delay.
The study suggests AI may relieve part of that bottleneck. By increasing the number of patients seen and the number of invasive procedures performed, the tool appears to have expanded access to specialty evaluation rather than merely adding another layer of software to the chart. Whether that expansion translates into more curative resections, fewer missed cancers, or better survival remains unanswered here. The present data stop at throughput, procedure counts, and diagnosis timing.
The next step is not more enthusiasm. It is external validation in broader systems, with attention to stage at diagnosis, downstream treatment, and patient outcomes, not just clinic volume. Until then, this study supports a narrower conclusion: AI-assisted lung nodule triage can move more patients through the pipeline, but it has not yet shown that it shortens the diagnostic journey or definitively increases early cancer detection across all settings.