MA

Osama Hamdy

Currently, the U.S. has only around 6,000 endocrinologists to treat 34 million patients with diabetes, with 39 percent of them above the age of 55 years. This means one endocrinologist to every 56,000 patients. Even worse, it is expected that a decade from now, we will suffer a severe and catastrophic shortage when aging endocrinologists retire, especially that another 86 million U.S. adults are considered to have pre-diabetes.

Although primary care physicians (PCPs) currently treat 90 percent of patients with diabetes, expert specialists are needed in face of increasing management options and increasing sophistication of the comprehensive diabetes management plan. The urgent question is that can machines effectively replace experienced endocrinologists in the future? If not, can it at least become an endocrinology copilot that supports PCPs in decision making?

In fact, no other disease will benefit more from the recent advances in artificial intelligence (AI) than diabetes. After being in that field for 40 years, having been in the forefront of diabetes technology, I can confidently say that most patients with diabetes will be perfectly treated by AI in the very near future. So, how will this occur? Computers will be provided with data from four major sources: patient input through cell-phone applications, electronic health records (HER) which include other medical problems and anthropometric information, preset lab data that includes fasting c-peptide, fasting insulin and fasting plasma glucose to calculate insulin sensitivity (HOMA-IR), and anti-GAD-65 antibodies plus genetic information including common genetic polymorphism and HLA typing. The computer will also be provided with the national/international guidelines from leading societies and clinical pathways from reputable diabetes centers like the Joslin Clinic.

Each clinical pathway is specific for a unique phenotype cluster. From all this data, the computer will place each patient in one of the commonly known 5 phenotype clusters as a prerequisite for precision diabetes management. Input of comprehensive data together with other historical data from large diabetes databases will provide a proper diagnosis, predict possible complications, outline clinical prognosis, and ultimately suggest a precise management plan or recommend a referral to a group intervention program.

Additionally, AI has the capacity to accurately interpret the relationship between diabetes and other medical problems. It is especially adept at detecting those related to diabetes like dyslipidemia, hypertension, hepatic steatosis, sleep apnea, and coronary artery disease, among many others more than the human brain can do. AI can also evaluate drug-drug interaction and suggest the best drug option, including a management pathway for each cluster, suggesting a primary and a secondary choice of medications.

“AI can also evaluate drug-drug interaction and suggest the best drug option, including a management pathway for each cluster, suggesting a primary and a secondary choice of medications.”

Furthermore, management plans will be more comprehensive, as it will not only suggest a follow-up plan or future labs, but also provide a precise structured nutrition plan with menus and snack list, exercise plan with demo, referral plan to other specialties, and valuable handouts specifically designed for this unique patient - guided by input and historical data that include major determinants of health. Since physiologic monitoring data can be easily streamed from continuous glucose monitoring (CGM), cellular-based scale and blood pressure machines, and fitness trackers, the computer will closely monitor progress and see trends in glycemic control and timely intervene to give suggestions while providing continuous feedback for positive reinforcement.

Cognitive behavioral support can be also provided through AI and can be associated with a reward mechanism to encourage patients to adhere to the suggested management plan. Based on symptoms, the computer may also suggest/order an EKG, lower extremely doppler, or refer patients for neurological evaluation. Non-mydriatic photo of the patient’s fundus will be automatically screened and spontaneously analyzed to filter patients for proper referral for eye care.  Although AI may do a fully automated job in the future in many common phenotype clusters like Mild Obesity Diabetes (MOD), Mild Age-Related Diabetes (MARD) and to some extent in Severe Insulin Resistance Diabetes (SIRD), it may have some limitation in automatically managing patients in Severe Autoimmune Diabetes (SAID) and severe Insulin Deficiency Diabetes (SIDD) phenotype clusters, who may need an endocrinologist’s input for insulin planning.

Even so, endocrinology AI-copilot may still be able to guide PCPs to reach the endocrinologist’s level in treating such sophisticated or high-risk patients. AI may be also successful in monitoring and guiding patients on hybrid-closed and closed loop artificial pancreas, since 2-way digital interaction will be the future norm.

Although direct physical examination and the traditional human touch and empathy will be lost and cannot be replaced by a machine, accurate diagnosis and precision management will have a major health-economic benefit that outweigh these drawbacks. I expect that at the beginning of the AI wave, there will be a hybrid model between fully automated and in-person traditional management of diabetes, where frequency and duration of visits are significantly reduced.

Considering diabetes is a major epidemic, the AI industry is expected to reach tens of billions of dollars in the coming 10 years. I can clearly vision the future now, and I see a future where I will be able to focus on the most severe and sophisticated cases while far more people are able to easily access important, personalized, diabetes care in ways that are never seen before.