Tempus is a Chicago-based AI company that develops precision medicine[a] solutions using patients’ multi-modal data. The company was founded in 2015 and now boasts the world’s largest library of clinical and molecular data. Tempus focused initially on fighting Cancer but is now expanding its solutions into infectious diseases and mental health.
Originally, Tempus had three main lines of business in a symbiotic cycle.
- Genomic sequencing – Tempus offers a broad range of DNA and RNA sequencing tests that Physicians can order for their patients. In this product, Tempus combines genomic data with clinical data to provide insights on gene alterations associated with specific types of tumor and enable oncologists to prioritize among different treatment options.
- Clinical data structuring – Tempus partners with health-care systems and academic institutions to collect, organize and aggregate clinical data primarily but not exclusively from electronic health records. The company uses image recognition, optical character recognition (OCR) and natural language processing (NLP) to transform unstructured information that is buried inside radiology scans, pathology slides, health reports and oncologists notes into structured data.
- Data monetization – Tempus sells curated and anonymous data to pharmaceutical companies for research and development of new drugs.
Earlier this year, Tempus announced a bold new addition to its product offering. The company launched the beta version of Tempus ONE, a voice-enabled portable smart-speaker designed for helping physicians on real-time. The device is intended to be a physical manifestation of the Tempus library. It can conveniently lie in a doctor’s desk and is small enough to be carried in the pocket.
After listening to a question, Tempus ONE searches the answer on the library, organizes the data into a clinical context, and makes it seamlessly available for doctors as they take care of patients. Tempus ONE lets doctors ask a broad range of inquiries, including questions about a patient’s condition, clinical and genomic information, and test status (e.g.: “Tempus ONE, what is Karla Sanders’ MSI status?”, “Tempus ONE, what alterations were found in Karla Sanders?”).
Watch the video below for Tempus ONE Demo:
While the advertised job-to-be-done of Tempus ONE is to answer doctors’ questions, it’s likely that Tempus intends to use the device to collect conversational data from doctor-patient interactions. This new category of data would certainly make Tempus’ library even more powerful.
Challenges and limitations
It’s true that the use of AI in healthcare enabled by large sets of historical data is very powerful and exciting. AI perhaps is the single most promising technology to improve people’s health by finding the cure for diseases, developing new drugs and vaccines, and improving therapies. However, a lot of challenges lay ahead. Below, I exemplify the most important ones:
- Limitations because of data privacy and restricted data sharing – Data privacy is increasingly a challenge for digital companies, and health data is even more sensitive. Not only are patients usually not comfortable that others know details about their health, especially when it comes to serious diseases like cancer, but also a patient’s data can be used harmfully against the very same patient who provided it. Think of what happens if insurance companies suddenly get access to data about customers with terminal diseases, for example. For those reasons, sharing of data among health institutions is very restricted, which limits research and development efforts. Every application that uses identified patient data has be carefully thought through under ethical, privacy and legal lenses.
- Issues with Data quality – Algorithms are only as good as the data. Inconsistences and flaws in the systems that collect, process and record data undermine data quality, which in turn makes algorithms weaker. Also, a lot of data is still not captured because it remains in analog means, like a doctor hand note or a physical x-ray sheets, or in a system that is digital yet not integrated as a source of data. Digitalization of analog data and systems integration are all subjected to failures and often add more errors to the data.
- Lack of data for new diseases – AI algorithms rely on historical data. For novel diseases like the COVID-19, too little data is available. It takes a long time until there’s enough information to train, test and validate AI models, long enough to say that it’s likely that a pandemic like this one will come to an end before the most actionable insights can reliably be drawn from the data.
[a] Wikipedia defines precision medicine (PM) as “a medical model that proposes the customization of healthcare, with medical decisions, treatments, practices, or products being tailored to a subgroup of patients, instead of a one‐drug‐fits‐all model.”