Building for Diabetes? 12 Things to Get Right With Your Health Data
A practical guide for teams working on type 2 diabetes and prediabetes: digital health products, clinics, pharma-supported programmes, insurers and employer health programmes.
Suhina Singh
Founder & CEO, Jonda Health
September 30, 2026
·
12 min read
Diabetes is one of the most measured conditions in medicine. Glucose is checked at home, in clinics and continuously through sensors, HbA1c is tracked every few months, and kidney, heart, eye and foot health are monitored for years. On paper, it should be the easiest condition to build good digital tools for. In practice, that volume of data is exactly what makes it hard. Readings arrive in different units, from different devices, under different definitions, and a tool is only as reliable as the way it brings them together.
The scale of the need makes getting this right urgent. According to the IDF Diabetes Atlas 2025, 589 million adults are living with diabetes, around 43% of them are undiagnosed, and more than four in five live in low and middle income countries. A further 635 million adults have impaired glucose tolerance, which places them at higher risk of developing type 2 diabetes.
Adults aged 20 to 79 living with diabetes, by IDF region. Source: IDF Diabetes Atlas, 11th edition (2025). The IDF South-East Asia region includes India; the Western Pacific includes China.
Asia carries the largest share. The Western Pacific and South-East Asia regions together account for more than half of all adults with diabetes. The Middle East and North Africa has the highest prevalence, with around one adult in six affected, and Africa faces the fastest growth, with numbers projected to more than double by 2050. These are also the regions where records are most likely to span many languages, laboratory systems and unit conventions, and where much of the data still arrives as paper, PDFs and photos.
We wrote this guide for teams working on type 2 diabetes and prediabetes. If you are building a digital health product, it sets out the data decisions to get right before you scale. If you run a clinic or chronic care service, it covers what it takes to keep records coherent across providers and years. If you lead a pharma-supported programme, it explains what your data needs in order to show outcomes and support real-world evidence. If you are an insurer or run an employer health programme, it shows how definitions and data quality shape who you reach and what you can demonstrate. Type 1 and gestational diabetes have their own considerations and are outside the scope of this guide.
Where diabetes data actually lives
A person with type 2 diabetes may check their glucose with a meter at home, wear a sensor for a few weeks, receive HbA1c, lipid and kidney results through a laboratory portal, and collect medicines from more than one pharmacy. Eye and foot screening results often arrive as separate reports. Their app logs meals and activity, and their GP, endocrinologist and perhaps a diabetes educator each keep their own notes. None of these sources were designed to work together.
Each source holds part of the story. The value comes from bringing them together in a way that stays accurate over years of care.
The twelve points below are grouped into four themes: getting the clinical picture right, making the data usable, connecting the longer story, and building for trust and scale.
Get the clinical picture right
1Know which definition you are using
Prediabetes does not have a single global definition. The American Diabetes Association uses a fasting glucose of 5.6 to 6.9 mmol/L (100 to 125 mg/dL) or an HbA1c of 5.7% to 6.4% (39 to 47 mmol/mol). The World Health Organization sets the fasting threshold higher, at 6.1 mmol/L (110 mg/dL), and does not use HbA1c to define intermediate hyperglycaemia. Many national guidelines across Asia, the Middle East and Africa follow one approach or blend the two.
The choice changes who a product or programme counts. The same population can produce a very different number of people at risk depending on which thresholds are applied, which matters for eligibility, reported outcomes and, for insurers and employers, cost projections. Problems appear when definitions are mixed silently, for example when a programme recruits on one standard and reports outcomes on another.
What good looks like: the definition and thresholds you use written down, applied consistently, and recorded alongside each classification so results can be compared or recalculated later.
Matters most to: StartupsInsurers and employersPharma programmes
2Know when HbA1c misleads
HbA1c reflects average glucose over roughly three months, but only when red blood cells behave as expected. Anaemia, recent blood loss or transfusion, chronic kidney disease and pregnancy can all distort it. Haemoglobin variants and thalassaemia matter too. These are common in parts of Southeast Asia, South Asia, the Middle East, Africa and the Mediterranean, and depending on the laboratory method, they can make HbA1c falsely high, falsely low or unreportable.
For a tool serving large populations in these regions, this is not an edge case. A product that relies on HbA1c alone to classify risk or track progress can misclassify people in exactly the communities where diabetes is growing fastest.
What good looks like: HbA1c stored with the laboratory method and relevant context (haemoglobin variants, anaemia, kidney function, pregnancy), with fasting glucose or CGM data used alongside it where HbA1c may be unreliable.
Matters most to: ClinicsInsurers and employersStartups
3Keep the context of each reading
A glucose value of 9 mmol/L means one thing two hours after a meal and something quite different first thing in the morning. Whether a reading was fasting, post-meal or random, the time it was taken, and whether the person was unwell or had just changed medication all shape how it should be read. Home readings entered into apps often lose this context, particularly when they are typed in later from memory or a paper logbook.
What good looks like: each reading captured with its timing and type, prompts that make context quick to add, and logic that treats a reading without context differently from one with it.
Matters most to: StartupsClinics
Make the data usable
4Harmonise glucose and HbA1c units
Glucose is reported in mmol/L across much of Europe, Australia, China and many other countries, and in mg/dL in the United States, India, Japan and much of the Middle East and Latin America. HbA1c appears as a percentage in some systems and in mmol/mol in others, the latter being standard in the UK and parts of Europe. Across Asia and Africa, it is common to see both conventions within a single person’s record as they move between laboratories, private and public providers, or countries.
Without harmonisation, trends break and alerts misfire. A value of 7 could be a normal-looking HbA1c percentage or a fasting glucose at the diagnostic threshold, and a unit mismatch can turn a stable patient into an apparent emergency, or hide a real one.
What good looks like: every result stored with its original value and unit, converted to a consistent unit for analysis, with the source laboratory preserved so any figure can be traced back.
Matters most to: StartupsClinicsPharma programmes
5Treat meter readings with care
Home glucometers remain the main source of glucose data for most people with type 2 diabetes worldwide, particularly where CGM is not affordable or reimbursed. Readings reach products in many ways: synced over Bluetooth, typed in by hand, copied from paper logbooks, or photographed from the meter screen. Each route has its own failure points. Manual entry introduces typos and misplaced decimals, device clocks drift or sit in the wrong time zone, and meters differ in accuracy and in whether they display mmol/L or mg/dL.
Photo capture is an important bridge in markets where connected meters are rare, but it needs to read the value, unit and timestamp reliably across many meter models, as we cover in How to Choose a Medical Device Reading API.
What good looks like: each reading stored with its capture method, device model, unit and a verified timestamp, with plausibility checks that catch obvious entry errors before they reach a clinician or an alert.
Matters most to: StartupsInsurers and employers
6Handle CGM data consistently
Continuous glucose monitoring has changed what is possible, and the 2026 ADA Standards of Care recognise time in range as a valid measure of glycaemic control alongside HbA1c. The international consensus on CGM metrics recommends a target range of 3.9 to 10.0 mmol/L (70 to 180 mg/dL), with more than 70% of readings in range for most adults with diabetes, based on at least 14 days of data with the sensor worn most of the time.
The data behind those metrics is less tidy than the dashboards suggest. Sensors have warm-up periods and gaps, vendors export in different formats, and some people wear a sensor for two weeks and then stop. If time in range is calculated differently across devices or programmes, or from too little data, comparisons between groups become unreliable.
What good looks like: CGM data imported from each vendor into one consistent structure, standard metrics calculated the same way every time, and sensor wear reported alongside every metric so readers know how much data sits behind it.
Matters most to: StartupsPharma programmesClinics
Connect the longer story
7Structure medication histories
Diabetes treatment changes over time and varies widely by country. Metformin, sulfonylureas, DPP-4 inhibitors, SGLT2 inhibitors, weekly GLP-1 receptor agonists and several types of insulin appear under many brand names, in fixed-dose combinations and at different doses. Weekly injectables, supply shortages and switches between brands add complexity, and some treatments carry a risk of hypoglycaemia that should shape how a tool responds to low readings.
What good looks like: each medication recorded with its active ingredients, dose, frequency, start and stop dates and prescribing source, mapped to medical coding standards, with medicines that carry a hypoglycaemia risk clearly flagged.
Matters most to: ClinicsPharma programmes
8Link kidney, heart, eye and foot
Much of the long-term harm of diabetes lies in its complications. Kidney function (eGFR and the urine albumin to creatinine ratio), lipids, blood pressure, retinal screening and foot examinations are all tracked over years, often by different providers. Each has its own reporting quirks: eGFR equations have changed over time, albuminuria can be reported in several ways, and screening results often arrive as PDFs or scanned forms.
A tool that tracks glucose well but leaves out this picture misses much of what determines a person’s long-term health, and misses the outcomes that payers and programmes most want to change.
What good looks like: complication markers held in the same longitudinal record as glucose and treatment, with the calculation method or report source kept alongside each value.
Matters most to: ClinicsPharma programmesInsurers and employers
9Use lifestyle data honestly
Food logs, step counts, weight and sleep are central to prevention and lifestyle programmes, and they are also the least reliable data in the record. People log inconsistently, food databases are often built around Western diets and miss regional dishes, and wearables estimate activity in different ways. The risk is drawing firm conclusions, such as that a meal caused a spike or that a programme reduced risk, from data that cannot support them.
What good looks like: lifestyle data labelled by source and completeness, food databases that reflect the cuisines of the people you serve, and outcomes reported with honest caveats about what the data can and cannot show.
Matters most to: StartupsInsurers and employers
Build for trust and scale
10Give clinicians summaries and clear escalation rules
People often arrive at appointments with months of readings, and clinicians have only minutes. A useful tool produces a concise summary: glucose patterns by time of day, CGM metrics with sensor wear, HbA1c and complication markers with dates and units, current treatment, and anything that needs attention.
Some readings should never simply be logged. Low glucose below 3.9 mmol/L (70 mg/dL), and especially below 3.0 mmol/L (54 mg/dL), very high readings, and symptoms that suggest a hyperglycaemic emergency need clear rules that direct the person to act or seek care.
What good looks like: an exportable clinician summary in a consistent format, escalation rules agreed with clinicians and tested, and a record that can move into a clinic’s own system.
Matters most to: ClinicsStartups
11Know when your tool becomes a medical device
Many diabetes products start as logging or coaching tools. The line into regulated software is usually crossed by what a product claims or does: calculating insulin doses, recommending treatment changes, diagnosing diabetes or prediabetes, or predicting an individual’s risk of complications. Regulators in Singapore, the UK, the European Union, Australia, the United States and many other markets treat these functions as medical devices, each with its own rules.
What good looks like: an intended purpose written down early, claims reviewed against the rules in each market you serve, and validation evidence gathered from the start rather than retrofitted.
Matters most to: StartupsPharma programmes
12Design for consent, privacy and portability
Diabetes data is health data protected under laws such as GDPR, Singapore’s PDPA and, where it applies, HIPAA in the United States. The stakes rise when insurers and employers are involved, because people need confidence that data shared for their care or wellbeing will not be used against them. Consent should be specific about how data will be used, and data used beyond direct care should be pseudonymised, with access controls and audit trails.
Diabetes is also lifelong, while products and programmes come and go. Structuring data to interoperability standards such as FHIR, with results and medications mapped to medical coding standards, keeps it useful when people change apps, clinics, insurers or employers.
What good looks like: granular consent recorded alongside the data, clear separation between programme data and employment or underwriting decisions, pseudonymisation for secondary use, and standard-format export with provenance intact.
Matters most to: Insurers and employersStartupsClinicsPharma programmes
Where to start, depending on who you are
You do not need to solve all twelve at once. The right starting point depends on what you are building and what your data needs to do.
Digital health teams
Start with 1, 4, 5 and 11
These shape your core logic, your data model and your claims, and they are far cheaper to get right before launch than after.
Clinics and chronic care services
Start with 2, 3, 7 and 10
Your challenge is making sense of results from many sources in a short consultation, and knowing when a number cannot be taken at face value.
Pharma-supported programmes
Start with 6, 7, 8 and 12
These decide whether your programme can show outcomes that hold up, and whether its data can be combined with other sources once it ends.
Insurers and employer health programmes
Start with 1, 2, 9 and 12
These decide who you identify, whether the numbers you report are comparable over time, and whether members trust you with their data.
How Jonda Health can help
Some of the twelve points are data problems, and some are questions of design, strategy and delivery. We support both, so you can come to us with the whole picture rather than splitting it across several partners.
JondaX
Make your data usable
Read the devices people actually use. Our Medical Devices module turns a photo of a glucometer or blood pressure monitor into a structured reading with its value, unit and timestamp, which is ideal where connected devices are rare.Items 3 and 5
Bring laboratory results together. HbA1c, glucose, lipids and kidney markers from PDFs, scans and laboratory files in any format or language, harmonised to consistent units and mapped to medical coding standards, with clinical review built in.Items 2, 4 and 8
Connect wearables and build the longitudinal record. Activity and sleep data from hundreds of devices in one schema, alongside results and medications, ready to share or export.Items 7, 9 and 12
Jonda Health Services
Design, build and prove your programme
Digital health strategy. Choosing definitions and thresholds, defining intended purpose and claims, and deciding what to build yourselves versus bring in.Items 1 and 11
Experience design. Logging that people keep doing, context capture that takes seconds, and clinician summaries that fit a real consultation.Items 3 and 10
Automation and digital enablement. Programme operations, member and patient communication, and responsible use of AI in diabetes support.Items 10 and 12
Insight and value. Outcomes, benchmarks and evidence that show insurers, employers and health systems what your programme is delivering.Items 6, 8 and 9
Jonda Health is ISO 27001 certified, and our platform is designed to comply with HIPAA, GDPR and PDPA.
Talk it through with us
Whether you are building a diabetes product, running a chronic care service, planning a programme or designing a prevention benefit, we can help at any stage, from shaping the experience and strategy to getting the data right and showing the value it delivers. You do not need to have everything figured out before we speak.
