Taking Alzheimer's Disease as an Example: An In-depth Analysis of the AI + Chronic Disease Business Model

[baoshi.rao]

This article uses Alzheimer's disease as a case study to deeply analyze the AI + chronic disease business model. Let's take a look~

Alzheimer's disease cannot be cured and can only be definitively diagnosed after the patient's death. Once contracted, the optimal treatment effects gradually diminish over time, and most patients deteriorate progressively after diagnosis.

Experts have detailed how people's abilities decline from normal functioning to the "stages" of late-stage Alzheimer's disease.

According to related research, the diagnosis of Alzheimer's disease relies on an elimination process, subjective analysis, and some laboratory tests. The elimination process involves memory assessments, evaluations of potential behavioral changes, laboratory tests, and brain imaging. Certain medications, such as Aricept, Namenda, and Excelon, can help patients manage memory loss, but they do not reverse the underlying issue—they only slow the progression and assist with daily living. However, new radiological studies suggest that AI may be key to early detection of Alzheimer's, potentially preserving some memory functions for up to six years.

Alzheimer's disease, as a common and idiopathic neurodegenerative disorder, severely impacts the physical and mental health of the elderly in modern society. Clinically, it manifests as progressive deterioration of cognitive and memory functions, declining daily living abilities, and various psychiatric symptoms and behavioral disturbances.

The general diagnostic criteria include insidious onset, progressive intellectual decline, significant memory and cognitive impairments, mild neurological deficits, and typical imaging changes. Currently, Alzheimer's diagnosis still relies on exclusion methods: first, a dementia diagnosis based on clinical symptoms, followed by a comprehensive analysis of medical history, disease progression, physical examinations, and auxiliary tests to rule out other causes (e.g., syphilis) before arriving at a clinical diagnosis of AD. However, definitive diagnosis depends on brain tissue pathology.

Currently, approximately 50 million people worldwide suffer from dementia, a number that nearly doubles every 20 years. Yet, so far, all clinical trials for drugs to reverse this disease have failed.

Data from the International Alzheimer's Association shows: In 2013, there were 44 million dementia cases globally, with 50%-75% being Alzheimer's patients. In 2015, 9.9 million new dementia cases emerged worldwide, averaging one new case every 3 seconds. Globally, the total cost of dementia care in 2015 was estimated at $818 billion, an increase of $214 billion from 2010.

By 2030, this number is expected to rise to 152 million. One person develops dementia every 3 seconds, and the current annual cost of dementia is estimated at $1 trillion, projected to double by 2050.

Currently, China has the highest number of Alzheimer's patients in the world. A 2014 survey revealed that 90% of Alzheimer's patients in China were undiagnosed and untreated.

While there is no method to halt the disease's progression in its late stages, evidence suggests that early detection allows Alzheimer's to be managed and controlled with medication, improving cognitive function and delaying clinical progression by 10-15 years.

In developed countries, the average annual cost per Alzheimer's patient is $33,000. Early diagnosis and intervention can significantly delay the need for nursing home care, saving an average of $10,000 annually. This represents a huge market and a significant challenge.

A key challenge for clinicians is that most patients are diagnosed when the disease is already advanced. Researchers hope new AI technologies can change this by providing early warning systems.

Alzheimer's is caused by the accumulation of sticky amyloid plaques and tau proteins, which disrupt neuronal communication. By the time symptoms like memory loss, mood swings, and personality changes appear, treatment becomes difficult.

Currently, treatment options for Alzheimer's are limited, and the best we can do is prevention.

Individuals can even adopt lifestyle changes that may delay the onset of Alzheimer's or prevent it entirely.

According to the Alzheimer's Association, delaying dementia onset by five years could halve the death toll, saving approximately 30,000 lives annually.

Patients typically undergo testing only after cognitive decline symptoms appear. However, the more we learn about Alzheimer's, the clearer it becomes that the disease begins much earlier than previously thought. Research shows that Alzheimer's starts damaging the brain over a decade before symptoms manifest. This may be the optimal time to begin treatment for the best therapeutic outcomes.

Currently, memory clinics in neurology, psychiatry, and geriatrics departments at major hospitals in China can diagnose Alzheimer's. Standard diagnostic procedures include medical history review, physical exams, dementia screening tests (scale assessments), neurological exams, and laboratory tests (e.g., blood tests, CT, MRI, PET/SPECT brain scans).

A common diagnostic tool is positron emission tomography (PET), which measures levels of specific molecules (e.g., glucose) in the brain to assess symptom severity. Glucose is the primary fuel for brain cells—the more active the cells, the more glucose they consume. As brain cells degenerate and die, glucose consumption gradually declines until it ceases entirely.

Other PET scans target proteins associated with Alzheimer's, but glucose PET scans are more common and affordable, especially in smaller medical facilities and developing countries, as they are also used for cancer staging. Radiologists often use these scans to detect Alzheimer's by identifying reduced glucose levels in the brain (particularly in the frontal and parietal lobes). However, because the disease progresses slowly, glucose changes are subtle and difficult to observe visually.

PET scan of an Alzheimer's patient's brain (Image source: National Institute on Aging)

With advancements in AI, scientists have explored combining machine learning with PET scans to diagnose early Alzheimer's more reliably.

Neurological diseases are highly complex, involving multiple disciplines. From childhood ADHD to elderly dementia, there is a vast unmet clinical need for rare and age-related diseases.

AI methods could have a significant impact, guiding patients toward the right treatment pathways and enabling earlier, better Alzheimer's diagnoses.

Given the growing elderly population, there is an urgent need for computer algorithms to enable precise diagnosis. Currently, no approved diagnostic tool for Alzheimer's exists, but the first targeted software is now in development.

In Shenzhen, China, a company called Brain Doctor (Yiwai Technology) focuses on AI-based Alzheimer's analysis. It has secured multiple rounds of funding, aiming to develop the world's first approved diagnostic tool for dementia.

Brain Doctor (Yiwai Technology) already employs machine learning to diagnose Alzheimer's, leveraging years of data from thousands of patients' EEGs and developing proprietary algorithms.

This collaboration in brain science will apply AI technology to more patients, fostering new software to accurately distinguish Alzheimer's from Lewy body dementia (LBD) via EEG analysis.

Why am I so familiar with this? Because I interviewed with this company!

Before the interview, I analyzed the entire chronic disease management solution in healthcare, so this article is more comprehensive.

The challenge with Alzheimer's is that by the time all clinical symptoms appear and a definitive diagnosis is possible, too many neurons have died, making reversal nearly impossible. If this algorithm can enable early diagnosis, neurologists could use it to identify memory decline as an Alzheimer's symptom sooner, helping patients access treatment faster.

The most evident sign of Alzheimer's disease is the accumulation of β-amyloid plaques in the brain. However, scientists have recently linked certain metabolic changes to the disease. Experts note that while people are adept at identifying specific biomarkers of the disease, metabolic changes represent a more global and subtle process.

Consequently, although doctors might observe signs of change in various images, it takes considerable time to map these changes and their subtle patterns across enough patients over a sufficient period to determine which changes predict the onset of Alzheimer's. This is precisely where artificial intelligence (AI) comes into play.

AI applied to medical imaging, primarily through deep learning, enables machines to analyze and interpret medical images. It serves as an auxiliary tool to assist doctors in diagnosis and treatment, helping them quickly obtain imaging information, perform qualitative and quantitative analyses, improve efficiency in image reading, and identify hidden lesions. AI accomplishes tasks such as lesion identification and annotation, 3D reconstruction, automatic target delineation, and adaptive radiotherapy through methods like image classification, object detection, image segmentation, and image retrieval. These applications span disease screening, diagnosis, and treatment stages.

Utilizing AI (especially machine learning) algorithms to analyze patient data such as scales and imaging and construct auxiliary diagnostic systems is one of the core tasks in current smart healthcare research and industrialization.

Deep learning, a recently emerging and highly effective technique in machine learning, performs representation learning on data, establishing learning methods inspired by the human visual cortex to interpret data. It has been widely applied to the processing and analysis of image, text, and speech data, achieving significant results.

Deep learning models possess efficient and powerful automatic representation capabilities, enabling doctors to quickly establish more scientific diagnostic models. Moreover, integrating with high-level medical institution resources can significantly enhance the dissemination of quality medical resources to primary healthcare facilities, improving their diagnostic capabilities.

Convolutional neural networks (CNNs) in deep learning require relatively fewer training samples and lower original image quality, making them suitable for research on various diseases where case collection is challenging. Currently, there is relatively little research on auxiliary diagnosis for Alzheimer's disease (AD) and behavioral variant frontotemporal dementia (bvFTD). We use structural magnetic resonance imaging (sMRI) data to build a CNN-based AD-bvFTD model for auxiliary differential diagnosis of AD and bvFTD to improve diagnostic accuracy.

I believe this algorithm has strong clinical relevance. However, before we can implement it, we need to validate and calibrate the algorithm in larger, more diverse patient cohorts, preferably from different continents and various settings. For example, using thin-slice high-resolution CT for pulmonary nodule detection instead of thick-slice data. On the basis of AI's potential applications, we should cover as many manufacturers, parameters, image qualities, and disease types as possible.

Testing and calibrating this algorithm on more diverse datasets, along with data preprocessing and building the AD-bvFTD classification model:

This algorithm begins to independently learn which features are important for predicting Alzheimer's disease diagnosis and which are not. Through training, comparison, testing, and final result analysis on the ADNI dataset, researchers found that the deep learning algorithm developed for early prediction of Alzheimer's achieved 82% specificity at 100% sensitivity and made correct disease predictions an average of 75.8 months—equivalent to six years—before the patient received a final diagnosis.

The article focuses on the profit model from the perspective of chronic disease management, which appears multiple times in our analytical framework. Chronic disease management can serve as a pain point in the healthcare industry to drive traffic, a means of monetizing traffic (as one of the service e-commerce models), and an important tool to enhance customer retention in monetization models such as pharmaceutical e-commerce and telemedicine. This is because chronic diseases are the main focus of mobile healthcare, and chronic disease management involves comprehensive health management for chronic disease patients throughout the entire healthcare process (including prevention, diagnosis, treatment, and rehabilitation), complementing other aspects. Mobile healthcare companies can build personalized big data through chronic disease management, enabling personalized continuous treatment and serving as an important means of retaining traffic in mobile healthcare.

Definition of chronic disease management: Centered on human health needs, it uses internet tools to provide comprehensive disease management services for chronic disease patients, including collecting relevant data and offering health and treatment recommendations based on the data. We define chronic disease management as a service. Simply selling wearable devices or medications is part of the related industry chain of chronic disease management services and falls outside our defined scope, making it difficult to benefit from the valuation of chronic disease management.

The industry is highly competitive, and companies that can integrate resources at low cost to form a closed loop may have an advantage. Any entity that can collect data and possess patient health data entry points may become a participant, with numerous potential competitors. The chronic disease management industry is fiercely competitive. Smart devices (vital sign data), pharmaceutical e-commerce (medication data), telemedicine (diagnostic and treatment data), doctor-patient platforms (diagnostic and treatment data), and patient medical record platforms (archival data)—any data collection and generation entry points—can enter chronic disease management and become potential competitors. Connected smart diagnostic devices collect data at the source, which can be used internally or directed to other platforms.

While this is the best way to collect data, the requirements for smart device R&D and funding are high. Companies lacking adaptability may face obsolescence after significant investments due to product upgrades, making data source control a double-edged sword.

There are eight major traffic entry points for acquiring users and collecting data: mobile medical devices, O2O platforms, consultation/appointment/health management system apps, family health service platforms, third-party testing centers, remote consultation centers, high-end new medical service institutions, neurology hospital groups, and nursing homes.

Smart devices (vital sign data), pharmaceutical e-commerce (medication data), telemedicine (diagnostic and treatment data), and patient medical record platforms (archival data) are common chronic disease management data entry points. Beyond these, the following may also become competitors.

Without insurance coverage, few entities offer mature closed-loop services, mostly focusing on a single环节. This is due to lack of policy support and the high investment required for a complete closed loop. Although there are many industry participants, few in China offer a full closed loop—most either collect data or analyze data, rarely doing both.

A closed loop refers to encompassing all core elements of chronic disease management (collecting accurate and continuous data, performing data analysis, and providing diagnostic and treatment feedback to patients). Most chronic disease platforms can only handle one环节, with others still expanding. This is largely due to the lack of payers for chronic disease management and the relatively high investment required.

Building all resources internally to create a complete closed loop requires significant investment. Collecting the most原始 data requires well-designed hardware, demanding high R&D and technical expertise. Developing medical-grade diagnostic devices is challenging without strong technical support and积累.

To achieve continuous data collection and diagnostic feedback, doctor supervision and assistance are essential. Without payers and sufficient doctor motivation, mobile healthcare companies must find suitable doctor resources and provide incentives, requiring substantial financial and effort investments. Telemedicine is currently limited to B2B, and electronic medical records are not yet personal products.

Currently, telemedicine cannot be directly implemented via B2C, meaning chronic disease management companies wanting to provide后端 treatment-level services must collaborate with offline medical institutions to offer完整的 services. Additionally, electronic medical records are not yet personal products, and information silos exist between medical institutions, resulting in incomplete personal data.

Under these circumstances, achieving a personal complete closed loop faces certain bottlenecks.  The insurance collaboration model is promising. Without insurance coverage, companies that can整合相关 resources at low cost to complete the闭环 are看好. With guaranteed product quality, those who gain insurance support first can批量 acquire users and医生 support, leading the industry.

The case study "Post-mortem of Elderly Care Entrepreneurship" from our company clearly illustrates this point. Therefore, in chronic disease management, we favor models that collaborate with insurance companies or governments, as having payers makes implementation easier. When insurance cannot serve as the payer, creating a data closed-loop becomes crucial for maximizing customer retention. The internet provides a relatively open platform where not all resources need to be built in-house - data and traffic can be mutually imported, as can physician resources.

Without medical insurance coverage, in capital-intensive, low-margin chronic disease management during its early stages, only those who survive long enough, accumulate sufficient active users, and persist till the end will ultimately succeed. With numerous chronic conditions having different pain points and profit models, the common denominator for successful self-pay models is: low-cost integration of relevant resources to complete the closed-loop and establish profitability.

Entities with technological or physician resources are more likely to become successful integrators. In China, who can integrate resources at low cost? As previously discussed, insurance involvement represents a superior profit model for chronic disease management - those who can collaborate with insurers will better integrate resources to form closed-loops.

Among existing self-pay models, which are more likely to succeed? Given the variety of conditions and profit models, it's difficult to predict winners. We examine potentially successful self-pay chronic disease management models by dividing them into two components: data collection and service provision through data feedback.

These components are interdependent. Some companies may gain first-mover advantage in data collection, aggregating large user bases that attract quality healthcare services for collaboration, while technological investments can reduce backend integration costs. Others may excel at service provision, attracting hardware companies to contribute traffic and saving R&D costs. Each component can become a breakthrough point, where we identify potential leaders.

In data collection, we favor "superior products + strong channels" or resource-rich healthcare companies. Three collection methods exist, each with potential leaders:

1. Direct patient data collection: Challenging; favors companies with leading product design and strong channels. Winners will benefit from C2B reverse bargaining power for resource integration. Methods include patient manual entry or sensor-based hardware-software combinations - likely the future mainstream given elderly patients' difficulty with manual entry.

2. Hardware-based collection: Requires advanced technology and sales channels. While low-end hardware faces intense competition, high-end solutions (accurate, user-friendly, cloud-integrated) have fewer competitors but need continuous technical and financial investment.

3. Future trends suggest unified data collection standards will emerge from current fragmentation. Market consolidation will favor companies with superior product design and distribution channels.

Data gatekeepers gain bargaining power over downstream services, but current homogeneous competition requires subsequent physician services to achieve closed-loop user retention.

Data integration without hardware faces intense competition. Success depends on providing better diagnostic services to create win-win scenarios with data collectors.

Physician-led approaches, leveraging doctors to promote hardware/apps and provide services, show stronger patient compliance. Given saturated patient-entry markets, physician/hospital-focused strategies may prevail, favoring companies with strong medical resources.

In service provision through data feedback, we favor companies with "medical-grade algorithms" and abundant healthcare resources. Sustained engagement requires precise, timely diagnostic services across four approaches:

1. AI-powered diagnostic suggestions (needing medical-grade algorithms and certification)
2. Teleconsultations (non-medical grade)
3. B2B telemedicine (currently B2C is prohibited)
4. Integrated offline medical resources

Smart suggestions can reduce labor costs and ease physician integration, but China lacks medical-grade AI standards. Consultations play crucial intermediary roles, where pharmacists and community doctors can effectively connect patients to specialists at lower costs. Companies rich in such resources gain advantages.

Integrating scarce, decentralized Chinese medical resources remains challenging but sticky. Companies capable of local healthcare resource integration will dominate chronic disease management.

The essence of chronic disease management lies in data management, encompassing the collection of accurate and continuous data, its analysis, and the provision of diagnostic or therapeutic recommendations based on the results. Any company capable of generating health data has the potential to enter the chronic disease management space, making the industry highly competitive. The question then becomes: who is most likely to emerge as the winner?

Under the condition of guaranteed product quality, those who first secure support from insurance providers or governments will be able to acquire users in bulk and gain physician endorsement, thereby leading the industry.

Thus, in chronic disease management, we favor models that collaborate with insurance companies or governments. With a payer in place, progress becomes significantly easier.

In scenarios where insurance cannot act as the payer, creating a data-driven closed loop is essential to maximize customer retention. We explore the common traits of potentially successful self-pay models: low-cost integration of relevant resources to complete the loop, a viable profit model, and the ability to foster strong customer loyalty.

From a model perspective, we anticipate the maturation of chronic disease management once insurance becomes involved. If public healthcare does not cover costs, self-pay models will require mobile chronic disease management companies to seek breakthroughs at various stages. Continuing from Section 3, we are optimistic about companies that can integrate diverse resources at low costs to form closed-loop patient management systems.

As for who is most likely to succeed, we believe internet-based enterprises, as open platforms, can attract and aggregate other resources to reduce costs—provided they excel in a specific niche. The key lies in leveraging their unique strengths to become industry leaders in their chosen segments.

Among the potential winning models we’ve identified, they ultimately fall into two broad categories: