Scalable Prognostics from Routine Clinical Data

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2105 papers123 conceptsfrozen 2026-10-04

Precision medicine is often associated with high-cost, high-complexity technologies like whole-genome sequencing or advanced functional imaging. However, a growing body of research suggests that the "hidden" signals required for accurate risk stratification already exist within the most mundane and ubiquitous clinical data: the standard blood panel, the 12-lead ECG, and routine fetal ultrasounds.

By applying machine learning and statistical ratios to these low-cost data sources, researchers are developing prognostic tools that are not only as accurate as their high-tech counterparts but are also immediately deployable in resource-limited settings.

The Power of Biological Ratios

The most common approach in this field involves calculating ratios between standard laboratory metrics. Individually, a neutrophil count or a lymphocyte count might stay within "normal" ranges, but their ratio (NLR) often captures a systemic state of inflammation or physiological stress that absolute values miss.

Wei Yan and colleagues demonstrated that the Neutrophil-to-Lymphocyte Ratio is a powerful predictor of cardiac death and rehospitalization in elderly heart failure patients. Similarly, Ruizhe Xu showed that the Platelet-to-Lymphocyte Ratio (PLR) serves as an independent risk factor for survival in cervical cancer patients undergoing radiotherapy.

These ratios work because they act as proxies for complex biological balances. For instance, the De Ritis ratio (AST/ALT), highlighted by Guangcheng Luo, leverages the different metabolic roles of liver enzymes to predict bladder cancer prognosis. By looking at the relationship between variables rather than isolated thresholds, clinicians can extract a higher-fidelity signal of systemic health from a standard metabolic panel.

Extracting Hidden Features with Machine Learning

While simple ratios provide a baseline, machine learning models can identify non-linear patterns across dozens of routine variables simultaneously. This transforms a standard hospital admission blood test into a sophisticated screening tool.

For example, Felipe Soares developed the ER-CoV tool, which uses 16 common blood variables to identify COVID-19 cases in emergency rooms. This model achieved a 95% negative predictive value, potentially reducing the need for expensive RT-PCR tests by 90%. In oncology, Wenhai Zhang utilized LightGBM to predict bone metastasis in breast cancer using only routine blood biomarkers, achieving an AUC of over 0.90.

The intuition here is that complex diseases like cancer or systemic infections leave a "fingerprint" across multiple physiological systems—affecting electrolytes, clotting factors, and immune cell distributions simultaneously. While a human clinician cannot easily synthesize 20 different lab values into a single risk score, models like the one developed by Jan Clusmann can process data from millions of individuals to predict liver cancer risk with high precision.

Repurposing Ubiquitous Waveforms and Biometrics

Beyond blood work, the same philosophy is being applied to common diagnostic waveforms and imaging. The 12-lead ECG is perhaps the most ubiquitous cardiac test in the world, yet much of its raw data is ignored in standard clinical interpretation.

John Theurer and Neal Yuan developed PreOpNet, a deep learning model that analyzes raw ECG waveforms to predict post-operative mortality. By identifying subtle morphological patterns in the electrical signal that are invisible to the naked eye, the model outperformed traditional clinical risk scores.

A similar logic applies to prenatal care. Christie L. Walker identified that a specific "femur-sparing" growth profile in routine ultrasounds—measured simply as the ratio of head circumference to femur length—could predict Zika-associated brain injuries more accurately than standard microcephaly screenings.

Why Routine Data Works as a Proxy

The success of these models rests on the fact that routine metrics are often downstream indicators of the same pathways targeted by expensive "omics" technologies:

  1. Inflammation and Stress: Markers like Red Cell Distribution Width (RDW), studied by Xiaofeng Liu and Meng Hao, reflect underlying oxidative stress and chronic inflammation.
  2. Metabolic Integrity: The Triglyceride-Glucose (TyG) index, explored by Y. Zhou, serves as a low-cost proxy for insulin resistance.
  3. Organ Reserve: Simple combinations of hemoglobin, bilirubin, and creatinine, as used in the "HTC model" by Shouhao Wang, can predict the onset of hepatorenal syndrome more effectively than specialized scores.

By treating standard clinical data as a rich, high-dimensional feature set rather than a collection of isolated values, this research direction provides a path toward equitable precision medicine. It suggests that the most powerful diagnostic tool in a clinic may not be a new genomic sequencer, but a better way of reading the data already being collected.

Papers in this version

Showing 60 of 105 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 67 papers
  2. Version 1 · 2026-10-04 · 105 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_biomarker_accessibility_b6567aab,
  title = {biomarker accessibility},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/biomarker-accessibility/rdr_b6567aabc4b34e36937e55811b79e05e},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_scalable_prognostics_from_routine_clinic_b6567aab_v1,
  title = {Scalable Prognostics from Routine Clinical Data},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/scalable-prognostics-from-routine-clinical-data/rdr_b6567aabc4b34e36937e55811b79e05e/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}