Copper Damage Leaves a Distinctive Blood Fingerprint

Disease Detection Series • Wilson Disease

Wilson disease, a genetic disorder of copper metabolism affecting approximately 1 in 30,000 individuals, is nearly always diagnosed after irreversible damage—cirrhosis, neurological devastation, or both. Yet it leaves a highly distinctive fingerprint on routine liver function tests for years before that damage becomes irreversible.¹ Treatment with copper chelation initiated before neurological involvement halts disease progression and provides near-normal life expectancy. The gap between available treatment and current diagnosis is not a gap in medical knowledge; it is a gap in pattern recognition.

Three in four Wilson disease patients receive at least one incorrect diagnosis before the disease is identified. The blood pattern that distinguishes it has been in their liver function tests throughout.

The Treatment Effect That Makes Early Detection Urgent

Wilson disease is uniformly fatal without treatment. Copper accumulation, present from birth because the ATP7B gene mutation impairs biliary copper excretion, eventually overwhelms hepatic compensation, triggering liver failure, chronic cirrhosis, or neuropsychiatric deterioration. Approximately 3 percent of all acute liver failure cases at transplant centers are attributable to Wilson disease.² Chelation with D-penicillamine or trientine initiated before neurological symptoms appear achieves stabilization or improvement in 70 to 80 percent of patients. Those treated after established neurological involvement have substantially worse outcomes.

The Signature in the Blood

Wilson disease produces a characteristic liver function pattern that distinguishes it from virtually every other liver disease. The critical discriminating feature is disproportionately low or normal alkaline phosphatase (ALP) in the setting of active hepatitis. In virtually all other liver diseases, ALP rises with hepatocellular injury; in Wilson disease, copper directly impairs ALP synthesis. Elevated bilirubin with low-normal ALP in a patient aged 5 to 35 years should trigger immediate evaluation.³

ALT (alanine aminotransferase) and AST (aspartate aminotransferase) are elevated, often the first and only abnormality for years. Albumin declines progressively as synthetic function deteriorates. Bilirubin rises initially in the direct fraction. Ceruloplasmin is low in most affected patients, though sensitivity is limited in isolation. Coombs-negative hemolytic anemia (identified on a routine complete blood count by declining hemoglobin and rising bilirubin) is present in acute presentations and is itself a red flag when accompanied by elevated liver enzymes.

The Machine Learning Case

Chen and colleagues applied machine learning to routine blood parameters in 346 Wilson disease patients to predict cirrhosis, identifying platelet large cell count, red blood cell distribution width-coefficient of variation (RDW-CV), ceruloplasmin, and mean corpuscular volume as key predictive features—all present on standard hematological and biochemical panels.⁴

Liang and colleagues applied XGBoost to 3,559 Wilson disease patients from the Optum real-world electronic health record (EHR), modeling disease progression using liver function tests and metabolic panel values. Dynamic area under the curve (AUC) reached 0.82 at one year for liver failure prediction, demonstrating that longitudinal patterns in standard labs can stratify Wilson patients by progression risk.⁵ The biological basis: copper-induced oxidative damage produces a trajectory of ALT/AST elevation, albumin decline, and bilirubin rise that follows a characterizable natural history—a trajectory machine learning reads better than any individual threshold.

What This Proves

Wilson disease is rare but treatable, and its blood signature is distinctive enough that a trained algorithm reading elevated transaminases, low-normal ALP, declining albumin, and rising bilirubin in a young patient could flag the diagnosis years before neurological symptoms arrive. No deployed population-level machine learning algorithm yet exists, but the biological and computational evidence for feasibility is established. The algorithmic gap is closable.

Endnotes

1.Czołnkowska A, Litwin T, Dusek P, et al. “Wilson Disease.” Nat Rev Dis Primers. 2018;4:21.

Comprehensive review covering prevalence, pathophysiology, and treatment of Wilson disease. Documents that chelation before neurological involvement achieves stabilization in 70–80%, and that delay causes irreversible damage. Primary reference for disease biology and treatment efficacy.

2.Roberts EA, Schilsky ML. “Diagnosis and Treatment of Wilson Disease: An Update.” Hepatology. 2008;47(6):2089–2111.

AASLD practice guideline documenting Wilson disease contribution to ~3% of acute liver failure cases and detailing the liver function pattern, particularly low ALP with elevated bilirubin, that characterizes hepatic presentation. Standard clinical reference for the diagnostic approach.

3.Cauza E, Maier-Dobersberger T, Polli C, Kaserer K, Kramer L, Ferenci P. “Screening for Wilson’s Disease in Patients with Liver Diseases by Serum Ceruloplasmin.” J Hepatol. 1997;27(2):358–362.

Prospective screening study documenting ceruloplasmin alone has limited PPV for Wilson disease in unselected patients, establishing the need for multi-marker approaches. The low-ALP-with-elevated-bilirubin combination is the key discriminating pattern that machine learning is designed to exploit.

4.Chen K, Wan Y, Mao J, Lai Y, Zhuo-ma G, Hong P. “Liver Cirrhosis Prediction for Patients with Wilson Disease Based on Machine Learning: a case–control study from southwest China.” Eur J Gastroenterol Hepatol. 2022;34(10):1067–1073.

Machine learning model on 346 Wilson disease patients using blood routine parameters to predict cirrhosis. Demonstrates that routine hematological parameters carry cirrhosis-predictive information in WD beyond what clinical thresholds capture, directly supporting the machine learning-based early detection argument.

5.Liang C, Kelly SP, Shen R, et al. “Predicting Wilson’s Disease Progression Using Machine Learning with Real-World Electronic Health Records.” medRxiv. 2023.

XGBoost on 3,559 WD patients from Optum EHR, achieving dynamic AUC 0.78–0.82 for liver failure prediction. The largest EHR-based WD machine learning study to date. Preprint pending peer review; results should be interpreted as proof-of-concept.