The scale of the analysis is one of its distinguishing features. Drawing on outpatient and inpatient records from Zhongshan Hospital between January 1, 2014, and November 5, 2024, the investigators assembled a cohort of 13,966 people with prediabetes. Laboratory test results and imaging examination data were collected from the hospital’s electronic systems, and the study endpoint was defined as the first occurrence of chronic kidney disease during follow-up. Because the cohort was so large and the observation window so long, the researchers had enough events to model risk not as a single static number but as something that evolves with time, which is precisely where their methodological choice becomes important.

That choice was the time-dependent Cox regression model, a statistical framework that extends the classical survival analysis approach of David Cox to accommodate predictor variables whose values change over the course of observation. In a conventional Cox model, a patient’s covariates are typically fixed at baseline, which can be a serious limitation in longitudinal medicine: blood pressure, medication regimens, lipid profiles, and cardiac function all drift over years, and a model that ignores that drift can misjudge risk. A time-dependent formulation allows the hazard of developing kidney disease to be recalculated as new clinical information accumulates, effectively letting the model update its forecast the way a weather model updates as fresh satellite data arrive.

Building such a model on more than a decade of real-world hospital records also posed a data-engineering challenge, and the team addressed it with a strikingly contemporary tool. During the research process, the authors report, DeepSeek-R1 was utilized for feature extraction from clinical data, an early example of a large language model being embedded in the pipeline of a clinical prediction study. Rather than hand-coding every relevant variable from free-text records and structured entries, the researchers used the model to help identify and organize the clinical features that would ultimately feed the survival analysis. The final predictor set that emerged is a portrait of the prediabetic patient as a whole-body system rather than a glucose value alone.

The variables incorporated into the model span several organ systems. Uric acid disorders, hypertension, and anemia each contributed predictive signal, as did low high-density lipoprotein cholesterol, the so-called good cholesterol whose protective vascular role is well established. Age entered the model per ten-year increment, and gender was included as well. Cardiac and vascular measures appeared in the form of left ventricular ejection fraction, a standard echocardiographic index of pumping efficiency, and carotid intima-media thickness, an ultrasound measurement of arterial wall thickening that serves as a surrogate marker of early atherosclerosis. The medication history proved informative in its own right: the number of distinct classes of glucose-lowering medications a patient was taking, whether the patient was on a renal-protective glucose-lowering agent, use of insulin, and use of novel oral anticoagulants all entered the final equation.

Each of these predictors tells a coherent biological story. Elevated uric acid is associated with endothelial dysfunction and renal injury, while hypertension imposes mechanical stress on the delicate filtering structures of the kidney. Anemia can both reflect and exacerbate renal impairment, since failing kidneys produce less erythropoietin. The finding that the intensity and type of glucose-lowering therapy carried predictive information suggests that treatment burden functions as a proxy for disease severity and trajectory, and the specific value of renal-protective agents echoes the growing clinical recognition that some modern drugs, such as SGLT-2 inhibitors, shield the kidneys directly rather than merely lowering blood sugar. Meanwhile, reduced ejection fraction and thicker carotid arteries tie kidney risk to the broader cardiovascular continuum, reinforcing the idea that the heart, the vasculature, and the kidneys fail together more often than in isolation.

To test whether the model actually worked, the researchers turned to five-fold cross-validation, a technique in which the dataset is split into five parts and the model is repeatedly trained on four of them and evaluated on the fifth, rotating through all partitions. This internal validation strategy guards against the most common failure mode of clinical prediction models: overfitting, in which an algorithm memorizes the quirks of its training data and then collapses when confronted with new patients. The performance metric was the time-dependent area under the curve, or AUC, which measures how well the model separates, at each time point, the patients who go on to develop kidney disease from those who do not.

The results showed a model that is strong in the near term and gracefully degrading over longer horizons. The time-dependent AUC values were 0.818 at one year, 0.815 at three years, 0.801 at five years, 0.782 at seven years, 0.744 at nine years, and 0.685 at nearly ten years, with confidence intervals reported for each estimate. In practical terms, an AUC above 0.80 is generally considered good discrimination, meaning the model correctly ranked the risk of two randomly chosen patients more than four times out of five during the first five years of follow-up. The authors characterized this as moderate but stable discriminative ability with satisfactory calibration over short- and intermediate-term follow-up, and they were explicit that the tool should serve as an early risk-screening reference rather than a definitive clinical decision-making instrument.

That framing matters, because the clinical stakes are enormous. Chronic kidney disease is a silent progression: kidney function can decline for years before symptoms appear, and by the time routine markers such as estimated glomerular filtration rate or urinary albumin-to-creatinine ratio cross diagnostic thresholds, much of the renal reserve may already be lost. Prediabetes, defined by impaired fasting glucose or impaired glucose tolerance, affects hundreds of millions of people worldwide according to estimates from bodies such as the International Diabetes Federation, and it represents the single largest identifiable reservoir of future diabetes and diabetic kidney disease. A screening tool that can flag, at the prediabetic stage, which patients are quietly heading toward renal failure would allow clinicians to intensify monitoring, optimize blood pressure and lipid management, and prioritize renal-protective therapies long before irreversible damage occurs.

The study, which was registered as a clinical trial under number ChiCTR2400089463 and approved by the Institutional Review Board of Zhongshan Hospital, Fudan University, also hints at where predictive medicine is heading. The combination of a classical survival framework with large language model-assisted feature extraction from electronic health records suggests a hybrid future in which decades-old biostatistics and cutting-edge artificial intelligence reinforce one another, each covering the other’s weaknesses. The authors are careful about the limits of their work: the model was internally validated in a single Chinese hospital cohort, and external validation in independent populations will be needed before broad deployment. Yet the core message stands on its own. For the vast population living in the gray zone between normal glucose and diabetes, kidney risk is not a distant abstraction but a measurable, modelable quantity, and with the right data, clinicians may soon be able to see it coming years in advance.

Subject of Research: Development and validation of a time-dependent Cox regression model predicting incident chronic kidney disease in prediabetic patients

Article Title: Development and validation of a risk predictive model for incident chronic kidney disease in prediabetic patients based on time-dependent Cox regression

Article References: Liu, P., Lan, C.-D., Jin, Y., Hu, B.-S., & Yang, H. (2026). Development and validation of a risk predictive model for incident chronic kidney disease in prediabetic patients based on time-dependent Cox regression. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02587-2

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02587-2

Keywords: prediabetes, chronic kidney disease, risk prediction model, time-dependent Cox regression, machine learning, electronic health records, uric acid, hypertension, anemia, carotid intima-media thickness, predictive medicine, cross-validation

The research team, led by Yamin Li and colleagues, analyzed 615 patients with pathologically confirmed papillary thyroid carcinoma who underwent total or near-total thyroidectomy with lymph node dissection followed by their first course of radioactive iodine ablation. The cohort included 189 men and 426 women ranging from 16 to 80 years of age, with a mean age of about 42 years. Using the 2025 American Thyroid Association recurrence-risk framework, the investigators re-stratified the patients into low/low-to-intermediate and intermediate-high/high risk categories, deliberately excluding anyone with distant metastases at baseline so that “high risk” in this cohort reflected only loco-regional disease features such as gross extrathyroidal extension, extranodal extension, bulky nodal disease, or aggressive histology.

Treatment response was assessed six months after iodine-131 therapy, a time point consistent with the ATA’s dynamic risk assessment system, using suppressed thyroglobulin, thyroglobulin antibody levels, diagnostic whole-body scintigraphy, and SPECT/CT imaging. Patients were classified into four categories: excellent response, indeterminate response, biochemical incomplete response, and structural incomplete response—the latter defined by suspicious imaging findings or biopsy-proven local or distant metastatic disease. For analytical purposes, the first three categories were grouped together as non-SIR. The contrast between risk strata was striking: structural incomplete response occurred in just 10.4 percent of low/low-to-intermediate risk patients but in 38.1 percent of the intermediate-high/high risk group, a difference the authors describe as highly statistically significant.

Digging deeper into each stratum, the researchers found that the determinants of treatment failure differed markedly depending on baseline risk. In the lower-risk group, univariate analysis flagged stimulated thyroglobulin, the presence of lymph node metastasis, thyroglobulin antibody levels, and the administered iodine-131 dose as significant factors, while gender, T stage, the number of lymph nodes removed, the lymph node ratio, and age showed no association. Receiver operating characteristic analysis identified optimal predictive cutoffs of 1.5 ng/mL for stimulated thyroglobulin and more than 2.5 metastatic nodes, and combining the two variables pushed the area under the curve to 0.768. Multivariable analysis in this group ultimately retained stimulated thyroglobulin, thyroglobulin antibody, and iodine-131 activity as factors associated with structural incomplete response.

The intermediate-high/high risk group told a different story. Here, stimulated thyroglobulin, the number of metastatic lymph nodes, the total number of nodes removed, the lymph node ratio, thyroglobulin antibody, and treatment dose all reached statistical significance on univariate testing. ROC-derived cutoffs were 11.625 ng/mL for stimulated thyroglobulin and 11.5 metastatic nodes, with the combination achieving an area under the curve of 0.786. Multivariable logistic regression distilled these down to three independent predictors: stimulated thyroglobulin, the number of metastatic lymph nodes, and the total number of nodes examined. Notably, each additional metastatic node raised the odds of structural incomplete response by 23 percent, while each additional node examined was modestly protective—an effect the authors attribute to more thorough surgical and pathological staging.

The most clinically consequential part of the study, however, focused on the intermediate-risk “grey zone,” the population in which the decision to administer radioactive iodine remains most contested. From 396 intermediate-risk patients—defined as those with recurrence risks between 10 and 30 percent under the 2025 framework—the team randomly partitioned 297 into a training set and 99 into a validation set. Five variables emerged as independent predictors of structural incomplete response in the training cohort: age, tumor size, the number of lymph node metastases, the lymph node ratio, and stimulated thyroglobulin. Each carried a biologically plausible signal. Older patients tend to have reduced radioiodine avidity; larger tumors reflect greater burden; and nodal metrics quantify the extent of metastatic disease.

From these five predictors the investigators constructed a nomogram, a point-based graphical calculator that clinicians can use at the bedside. ROC analysis supplied practical thresholds: more than 8.5 metastatic lymph nodes, stimulated thyroglobulin above 7.46 ng/mL, a lymph node ratio exceeding 0.30, age over 42.5 years, and tumor size greater than 1.05 cm. A physician scores each factor, sums the points, and reads off the patient’s individualized probability of structural incomplete response. In the training cohort the model achieved an area under the curve of 0.865, and bootstrap internal validation with 1,000 resamples and Harrell optimism correction yielded an optimism-corrected C-statistic of 0.853 with a calibration slope of 0.92—evidence of minimal overfitting. Variance inflation factors all fell below 1.1, and the two nodal variables were only weakly correlated, supporting the retention of both.

Discrimination slipped to 0.733 in the held-out validation set, though the authors note the confidence intervals overlap with the training estimate. Decision curve analysis added a further layer of reassurance: across threshold probabilities from 1 to 95 percent, the nomogram’s standardized net benefit exceeded both the treat-all and treat-none strategies, meaning that using the model to guide decisions would, in theory, improve clinical outcomes compared with indiscriminate approaches. Calibration curves in both cohorts showed good agreement between predicted and observed event rates, and the events-per-variable ratio of 12 met accepted standards for logistic regression modeling.

The findings dovetail with a growing body of literature on thyroglobulin dynamics and nodal burden. Prior work has shown that when stimulated thyroglobulin stays below 1 ng/mL, structural incomplete response is essentially never observed; between 1 and 10 ng/mL it occurs in fewer than 2 percent of patients; and above 10 ng/mL the rate climbs to more than 40 percent. Because thyroglobulin is produced only by thyroid tissue and its metastases, rising levels signal residual or recurrent cellular activity. Similarly, the lymph node ratio—positive nodes divided by nodes removed—has repeatedly been linked to poorer disease-specific and overall survival, with a cutoff near 0.3 emerging in earlier studies as prognostically meaningful. The Soochow team’s threshold of 0.30, with 50 percent sensitivity and 93 percent specificity, aligns closely with that precedent.

The authors are candid about the study’s limitations. It was retrospective and single-center, raising the specter of selection bias; the administered iodine-131 activity partly reflects disease severity rather than an independent cause of outcome; dichotomizing continuous predictors discards information; and, crucially, the cohort comprised only patients already selected for radioactive iodine, so the nomogram predicts response within treated patients rather than informing whether therapy should be given at all. External multicenter validation is required before routine clinical use. Still, the contribution is clear: by comparing the determinants of structural incomplete response across ATA risk strata and integrating them into a calibrated, decision-analytically validated tool, the study gives clinicians a way to bring quantitative precision to one of thyroid oncology’s most stubborn gray areas—potentially sparing lower-risk patients unnecessary radiation while intensifying surveillance for those most likely to harbor residual disease.

Subject of Research: Development of a nomogram predicting structural incomplete response to radioactive iodine therapy in intermediate-risk papillary thyroid carcinoma

Article Title: Prognostic Factor Analysis for Risk‐Stratified Papillary Thyroid Carcinoma and Nomogram Development for Predicting Structural Incomplete Response to Radioactive Iodine Therapy in Intermediate‐Risk Patients

Article References: Li, Y., He, Z., Zhao, M., & Zhang, B. (2026). Prognostic Factor Analysis for Risk‐Stratified Papillary Thyroid Carcinoma and Nomogram Development for Predicting Structural Incomplete Response to Radioactive Iodine Therapy in Intermediate‐Risk Patients. Cancer Reports, 9(9), Article e70692. https://doi.org/10.1002/cnr2.70692

Image Credits: AI Generated

DOI: 10.1002/cnr2.70692

Keywords: papillary thyroid carcinoma, radioactive iodine therapy, structural incomplete response, nomogram, thyroglobulin, lymph node metastasis, lymph node ratio, ATA risk stratification, thyroid cancer prognosis, risk prediction model, iodine-131 ablation, decision curve analysis

The retrospective study analyzed specimens from inpatients admitted between January and December 2023. From 3,151 clinical specimens, the laboratory recovered 1,860 non-repetitive bacterial strains. After applying rigorous deduplication criteria—retaining only the first isolate per patient per infection site within seven days, unless the antibiogram changed significantly—the team assembled a patient-level dataset of 1,045 unique individuals. Six bacterial species dominated the sample: Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus, Enterococcus faecium, and Enterococcus faecalis, together accounting for 70.1 percent of all non-repetitive isolates. Of the 1,045 patients, 253—24.21 percent—had infections caused by multidrug-resistant organisms, while 792 did not. Bacterial identification and antimicrobial susceptibility testing were performed using the VITEK2 compact automated system, with results interpreted according to the Clinical Laboratory Standards Institute M100-S31 breakpoints.

The baseline comparisons revealed telling differences between the two groups. Patients with multidrug-resistant infections were more often aged 60 or older (64.03 percent versus 56.82 percent), more frequently ESBL-positive (45.85 percent versus 20.96 percent), and more likely to be infected with S. aureus (29.25 percent versus 8.08 percent) or E. faecium (9.49 percent versus 1.01 percent). Interestingly, the multidrug-resistant group included fewer intensive care unit patients (17.00 percent versus 29.04 percent) and fewer respiratory tract infections, but more urinary tract infections (17.00 percent versus 8.84 percent), more patients from general surgical departments, and more from burn surgery. Sex distribution was similar between groups, and E. coli showed no significant difference. These patterns underscore that resistance risk is not confined to intensive care—it threads through surgical wards, burn units, and urology services alike.

Multivariable logistic regression distilled six independent predictors of multidrug-resistant infection: age of 60 years or older, urinary tract infection, ESBL production, P. aeruginosa, S. aureus, and E. faecium. The effect sizes varied dramatically. E. faecium carried the strongest association, with an odds ratio of 55.48 (95 percent confidence interval, 21.66 to 142.15), followed by S. aureus at 29.23 (16.41 to 52.07) and ESBL production at 15.18 (9.04 to 25.51). P. aeruginosa tripled the risk (odds ratio, 3.66), while urinary tract infection (odds ratio, 1.79) and older age (odds ratio, 1.47) contributed more modestly. Each predictor was assigned a score based on its regression coefficient, and the summed total projects onto a probability scale—turning a statistical model into a bedside-usable chart.

Performance metrics were encouraging. Internal validation with 1,000 bootstrap resamples yielded a corrected area under the receiver operating characteristic curve of 0.82 (95 percent confidence interval, 0.79 to 0.85), indicating good discrimination. The model achieved a sensitivity of 85 percent and a specificity of 70 percent, with a negative predictive value of 94 percent—a particularly valuable property, because a low score can help clinicians reasonably rule out multidrug resistance and avoid unnecessary broad-spectrum therapy. Calibration was excellent: the bootstrap-corrected calibration slope was 1.000 with an intercept of 0.000, the Hosmer-Lemeshow test returned a P value of 0.926, and the Brier score of 0.136 fell well below the 0.25 threshold for acceptable predictive accuracy. Decision curve analysis showed positive net benefit across threshold probabilities of roughly 20 to 45 percent, the range where antimicrobial decisions are most consequential.

The authors were notably careful about a subtle statistical pitfall: incorporation bias. Because ESBL positivity was included as a predictor while the outcome—resistance to at least three antibiotic families—is related to ESBL status, the team ran a sensitivity analysis excluding ESBL from the model. The area under the curve dropped from 0.82 to 0.70, confirming that ESBL contributes substantially to discrimination. Intriguingly, P. aeruginosa lost significance without ESBL adjustment (odds ratio falling to 0.74), suggesting its apparent effect was partially mediated by ESBL status, whereas S. aureus and E. faecium remained strongly significant, demonstrating that their predictive power is largely independent of the ESBL variable. The authors also emphasize that ESBL positivity does not equal multidrug resistance—many ESBL-producing isolates remain susceptible to aminoglycosides, fluoroquinolones, and carbapenems—which is precisely why a continuous probability estimate adds value beyond a binary ESBL result.

Robustness checks extended further. Firth penalized logistic regression, a technique that reduces small-sample bias for rare events, was applied because E. faecium isolates were sparse (only 47 isolates, of which 38 were multidrug-resistant). The penalized odds ratio for E. faecium, 51.40 (95 percent confidence interval, 21.24 to 135.52), closely matched the primary estimate. A patient-level sensitivity analysis confirmed that all six risk factors remained directionally and statistically consistent after deduplication. Notably, age failed to reach significance in the isolate-level data (P = 0.07) but became significant once repeated specimens were removed (P = 0.027), suggesting that multiple cultures from the same patient can dilute true risk signals—a methodological lesson with implications well beyond this single study.

The clinical logic of the tool is grounded in established biology. ESBLs are enzymes that inactivate most penicillins, cephalosporins, and related agents, and their encoding genes frequently travel with additional resistance mutations. Elderly patients face elevated risk through immunosenescence, frailty, and multimorbidity. Urinary tract infections are among the most common infections associated with resistant organisms, particularly when broad-spectrum antibiotics are prescribed empirically without urine culture. S. aureus in this cohort was predominantly recovered from burn wound secretions, while E. faecium—a gastrointestinal commensal turned opportunistic pathogen—appeared most often in urine and bile cultures from hepatobiliary surgery and urology patients. P. aeruginosa, a Gram-negative aerobe notorious for hospital-acquired pneumonia, was mostly isolated from sputum of intensive care patients, and its association with high mortality makes accurate early risk assessment especially consequential.

Important caveats temper the enthusiasm. This was a retrospective, single-center study at a county-level hospital, so the findings may reflect regional epidemiology that does not generalize elsewhere. Several potential confounders—prior antibiotic exposure, invasive devices, comorbidities, and immunosuppression—could not be fully captured, although an E-value analysis (E-value of 110 for E. faecium) suggests unmeasured confounding is unlikely to explain the strongest associations. More than 500 rare bacterial species were excluded for statistical stability, so extrapolation to uncommon pathogens should be cautious. The model’s prediction time point is also specific: it is designed for the window after species identification and ESBL phenotype are known but before full susceptibility results return, not for purely empirical decisions made before cultures are drawn.

For now, the authors position the nomogram as a supplementary reference within comprehensive clinical judgment, not a replacement for it. They call for prospective, multicenter external validation before widespread clinical implementation, and future versions may incorporate richer confounder data. Still, the study demonstrates that meaningful resistance prediction does not require academic medical centers or machine learning black boxes—it can emerge from careful, well-deduplicated patient-level data in a county hospital, using variables any microbiology laboratory already reports. As antimicrobial resistance continues to climb globally, tools that convert routine microbiology into early, individualized risk estimates could become a quiet but powerful ally in antibiotic stewardship, helping clinicians reserve last-line drugs for the patients who genuinely need them.

Subject of Research: Development and internal validation of a nomogram for predicting multidrug-resistant organism infections in a county-level hospital

Article Title: A practical nomogram for predicting multidrug-resistant organism infection in a tertiary county-level hospital to guide antimicrobial therapy

Article References: Han, L., Zhao, H., Cheng, J., & Gao, Y. (2026). A practical nomogram for predicting multidrug-resistant organism infection in a tertiary county-level hospital to guide antimicrobial therapy. New Microbes and New Infections, 74, Article 101853. https://doi.org/10.1016/j.nmni.2026.101853

Image Credits: AI Generated

DOI: 10.1016/j.nmni.2026.101853

Keywords: multidrug-resistant organisms, nomogram, antimicrobial resistance, ESBL, risk prediction model, antibiotic stewardship, county-level hospital, Enterococcus faecium, Staphylococcus aureus, urinary tract infection, logistic regression, infection control