Machine Learning for Acute Hepatic Porphyria Screening
Summary and Importance of Disease
A debilitating and potentially life-threatening family of genetic diseases, acute hepatic porphyria (AHP) is a family of rare diseases where a diagnosis is commonly delayed or completely absent. Affecting women at five times the rate of men, it has many “common” symptoms, such as anxiety, nausea, abdominal pain, and fatigue, that can present as other diseases. As a result, diagnosis can be delayed an average of 15 years.1 Living with AHP can drastically impact daily functioning and one’s quality of life. Furthermore, acute AHP attacks are often severe and require hospitalization.2 New research is being conducted to use machine learning to apply population-level informatics to help facilitate the diagnosis of AHP and other rare diseases, reducing diagnosis delay.3
Genetic Basis/Pathophysiology
AHP is a family of related metabolic disorders caused by enzyme activity deficiencies in the liver.4 Depending on the defect in one of eight enzymes, different porphyria can present, including acute intermittent porphyria (AIP), variegate porphyria (VP), aminolevulinic acid dehydratase deficiency porphyria (ALAD), and hereditary coproporphyria (HCP). Each subtype of AHP is attributed to specific genes affected, but are all similar in that AHP patients are predisposed to an accumulation of porphyrins (eg. aminolevulinic acid (ALA) and porphobilinogen (PBG)) in the liver.4 An enzyme that controls the heme pathway (the mechanism to synthesize hemoglobin, a protein necessary for binding oxygen in the bloodstream), ALAS1, is affected by the genetic mutation and accumulates. When the levels of porphyrins within the pathway are too high, toxicity to tissues may occur, manifesting as the aforementioned symptoms. Specifically, ALA and PBG are harmful to nerve cells and are responsible for AHP attacks, causing tachycardia, abdominal pain, vomiting, hallucinations, seizures and even coma.5 Genetic mutations causing AHP are usually autosomal dominant and have low penetrance, with families carrying the gene not having many, or any, affected family members. As a result, family history is a poor diagnostic tool.6
Past Research
AHPs are difficult to diagnose because of infrequent occurrence, general presentations, and genetic components which require specialized testing. AHP can be diagnosed with a urine test indicating levels of ALA, PBG, and other porphyrins, but its rarity and common symptoms prevent healthcare professionals from suspecting it.5 Usually, AHP is treated with symptom management and attack-trigger avoidance, although new treatments are being investigated. Panhematin (hemin for injection) is used for recurrent AIP attacks by replenishing the body’s heme and signalling it to inhibit the production of porphyrins upstream of heme.5 Givlaari was recently approved as a drug to reduce levels of ALAS1 protein, decreasing porphyrins and reducing attacks.5
The gap in knowledge occurs in the timely diagnosis of AHP. Electronic health record (EHR) data has been used all around the world to detect rare diseases such as cardiac amyloidosis, lipodystrophy, and more. A study by Cohen et al. (2020) aimed to employ machine learning of EHR to facilitate faster identification of AHP.3
Current Research and New Findings
In a study published in PLoS One, Cohen et al. (2020) used EHR data to create and train a machine learning algorithm that was applied to the dataset and ranked patients on their likelihood of being diagnosed with AHP.3 Investigators used approximately 200,000 patient records from 2009 to 2019 and developed a gold standard for the algorithm by manually reviewing charts for patients with a confirmed diagnosis of AHP. The data was then deconstructed to be used for statistical analysis and training of the machine learning model. By the end of the study, the model was trained with 141 features associated with AHP and was applied back to the dataset. Although there were 30 positive cases within the 200,000 patients, many more patients without an AHP diagnosis were identified as “AHP diagnostic testing likely indicated”. These were then confirmed with manual chart reviews by healthcare providers.
Some of the strongest positive predictors in the model included unexplained abdominal pain, pelvic and perineal pain, nausea and vomiting, and numerous pain and nausea medications. Ultimately, the study revealed four undiagnosed patients who likely have AHP, and 18 others who would likely benefit from testing. A follow-up study will be conducted to contact these patients and confirm the diagnosis.
Conclusion
With the valuable information that was gained from Cohen et al.’s study and future studies, the delay in diagnosis of AHP can greatly decrease. Early diagnosis will tremendously help patients with symptom-management, as well as prevent emotional, physical, and financial suffering. Rather than living for multiple years with a mystery disease, those affected will be able to proactively improve their quality of life.
Machine learning models can be applied to other diseases, especially rare diseases that physicians often overlook. The increase in digital medicine and the use of statistics and machine intelligence has a place within population-level health, consequently impacting the health of greater society. It allows healthcare professionals and decision-makers to be proactive, rather than reactive.
Anna Xia
Works Cited
1. Chen B, Solis‐Villa C, Hakenberg J, Qiao W, Srinivasan RR, Yasuda M, Balwani M, Doheny D, Peter I, Chen R, Desnick RJ. Acute intermittent porphyria: predicted pathogenicity of HMBS variants indicates extremely low penetrance of the autosomal dominant disease. Human mutation. 2016 Nov;37(11):1215-22.
2. Acute Hepatic Porphyria. MyGiHealth. https://mygi.health/education/diseases/porphyria (accessed March 12, 2021).
3. Cohen AM, Chamberlin S, Deloughery T, Nguyen M, Bedrick S, Meninger S, Ko JJ, Amin JJ, Wei AJ, Hersh W. Detecting rare diseases in electronic health records using machine learning and knowledge engineering: Case study of acute hepatic porphyria. PloS one. 2020 Jul 2;15(7):e0235574.
4. Kothadia JP, LaFreniere K, Shah JM. Acute Hepatic Porphyria.
5. Acute Intermittent Porphyria. NORD (National Organization for Rare Disorders) 2020. https://rarediseases.org/rare-diseases/acute-intermittent-porphyria/ (accessed March 12, 2021).
6. Pischik E, Kauppinen R. An update of clinical management of acute intermittent porphyria. The application of clinical genetics. 2015;8:201.
Cite This Article:
Xia A., Patel M., & Bhans M. Machine Learning for Acute Hepatic Porphyria Screening. Illustrated by Z. Hasan. Rare Disease Review. March 2022. DOI: 10.13140/RG.2.2.12884.48001
