Chest X-Ray Score Distinguished Acute Respiratory Distress Syndrome

Evidence

A chest X-ray score distinguished acute respiratory distress syndrome—a severe lung-injury syndrome—from other classifications in a selected intensive-care cohort. The lung-imaging study, published October 21 reanalyzed a prospective, single-center observational cohort of 236 adults receiving breathing support through a tube for at least two days. Of these, 92 met the Berlin diagnostic definition of the syndrome.

At least two experienced readers, blinded to clinical information and syndrome status, scored X-rays from intensive-care days 1 and 2 using the Radiographic Assessment of Lung Edema score, which assesses fluid buildup. Researchers averaged readers’ ratings and selected each patient’s higher-scoring day. The area under the receiver operating characteristic curve—a measure of separation between cases and noncases, where 1 represents perfect separation—was 0.93 (95% confidence interval, 0.90–0.96). At a cutoff of 20, the score identified 85% of cases and correctly classified 83% of noncases, with reported intervals of 76–91% and 76–89%, respectively.

The cirrhosis study, published October 6, retrospectively examined 416 patients with cirrhosis, or advanced liver scarring: 135 had bleeding from enlarged veins in the esophagus or stomach, and 281 did not. A machine-learning model combining nutritional and laboratory measures reported an area under the curve of 0.860. Among eight nutritional assessment tools, the Malnutrition Universal Screening Tool scored 0.698; the other indicators were below 0.6. Higher screening-tool scores were associated with bleeding in an analysis without adjustment for other factors: the odds ratio was 1.444 (95% confidence interval, 1.222–1.705), with values above 1 indicating higher odds.

The MEDA-PKD registry study, published October 20, examined adults with autosomal dominant polycystic kidney disease, an inherited kidney disorder. Its platform displayed data from 1,735 patients, with mean follow-up of 796 days. In 332 people starting tolvaptan, a statistical model accounting for repeated measurements found that estimated kidney filtration declined by 3.62 before treatment versus 3.01 during treatment, measured in milliliters per minute per 1.73 square meters of body surface area each year. The estimated slowing was 0.61 in those annual units (95% confidence interval, 0.07–1.16). Treatment initiation was also associated with an acute filtration drop of about 2.5 milliliters per minute per 1.73 square meters (95% confidence interval, 1.1–3.9).

Analysis

Interpretation: These studies measure different things: diagnostic separation in lung injury, model-based classification of bleeding in cirrhosis, and treatment-associated kidney-function trajectories. Their performance numbers are not directly comparable across populations and outcomes. The stronger performance of the combined cirrhosis model does not isolate the contribution of nutrition measures. Likewise, the kidney registry’s slower annual decline and initial filtration drop describe different aspects of the recorded trajectory, not a demonstrated causal treatment effect.

Limitations

The lung analysis was post hoc—an after-the-fact analysis—and excluded patients with trauma or heart-related lung-fluid buildup. Its averaged expert readings and selection of the higher-scoring day may not translate to one routine admission reading.

The cirrhosis cohort was retrospective, with 416 patients selected from 647 cases. These results do not establish independent validation or quantify uncertainty around the model’s discrimination. The kidney registry was observational, so treatment selection and other differences could influence the association. Portability of its platform also depends on structured hospital data, compatible systems and data-access rules. Although it integrates proteomics, or large-scale protein measurements, no protein-specific outcome estimate is reported. Statistical significance and classification performance do not themselves establish clinical benefit.

What to watch

Independent validation of the cirrhosis model and a suitable comparison group for kidney-function trajectories would address separate uncertainties. For the lung score, confidence would rise if a prospective multicenter study, using a prespecified cutoff of 20 and reading protocol, reproduced roughly 85% case detection and 83% correct noncase classification in a broader population, including patients with heart-related fluid buildup. Materially lower performance would weaken confidence in its generalizability.