.png)
The value proposition of a preventive medicine platform is specific and powerful: to detect early, in the user's health data, the signs of increased risk of a clinical event before that event happens. To identify the trend of rising blood glucose before the diabetes diagnosis. To recognize the deterioration of renal function before significant filtration loss. To flag the lipid pattern that precedes the cardiovascular event. It is a promise of prediction, and it is what differentiates preventive medicine from simple test storage.
For that promise to hold, the platform needs something most don't have: genuinely comparable biomarker time series, over months and years, in populations that can reach tens of thousands of users. Without that, what the platform calls prediction is, in practice, just the presentation of numbers that don't mean what it claims.
The difference between measuring and predicting
Measuring is recording a biomarker's value at one moment. Predicting is recognizing, in that biomarker's trajectory over time, a pattern that indicates future risk. The distinction looks subtle, but it is the difference between a real preventive medicine platform and one that only presents itself as such.
Prediction depends on longitudinal comparability. A creatinine value of 1.3 mg/dL means nothing in isolation from a predictive standpoint. It only gains meaning when compared with the same user's prior values: if creatinine had been stable at 0.9 mg/dL and rose to 1.3, that is a deterioration signal warranting attention; if it had oscillated between 1.2 and 1.4 for two years, it is stability. The platform's predictive capacity is entirely contained in its ability to reliably compare the current value with the prior trajectory.
And that is exactly where most platforms fail. When the user's tests come from different laboratories over time, which is the rule and not the exception, the values arrive with distinct nomenclatures, units that may vary, different analytical methods, and reference ranges specific to each laboratory. The platform that doesn't harmonize this data before building the time series is comparing values that are not comparable. And when the comparison isn't valid, the prediction built over it isn't prediction: it is noise presented with the appearance of signal.
When laboratory variability becomes a false clinical signal
The most dangerous failure mode of a preventive medicine platform without harmonization is the confusion between two types of variability: the patient's real variability, which is what the platform should detect, and the laboratory's methodological variability, which is an artifact the platform should eliminate.
Consider glycated hemoglobin, a central biomarker in metabolic risk monitoring. It can be reported in two distinct unit systems, NGSP/DCCT as a percentage and IFCC in mmol/mol, which produce completely different numbers for the same physiological reality. The literature documents the long standardization effort for this biomarker and the diagnostic implications of differences between methodologies.¹ A platform that receives a test reported in one system and the next test in another, without identifying and harmonizing the difference, will detect a "variation" in the user's biomarker that corresponds to no real physiological change. It will fire a risk alert that is pure artifact. Or, in the opposite direction, it will mask a real deterioration because the methodological difference pushed the number the other way.
Differences in reference ranges between laboratories worsen the problem. The literature demonstrates that variations in the reference intervals used by different laboratories can lead to diagnostic discrepancies in a significant proportion of cases for certain tests.² A platform that classifies a result as "normal" or "altered" based on the reference range that came in the report is applying ranges that vary from laboratory to laboratory, which means the same platform can classify the same value in opposite ways depending on where the test was done.
The result is a platform that produces alerts, trend charts, and risk classifications, all with the appearance of sophisticated preventive medicine, while the relationship between those outputs and the user's real clinical risk remains unknown. The platform is measuring. It is not predicting. And the difference is precisely the value proposition it sold.
The consequence for the business, not only for the user
For a preventive medicine healthtech, the inability to predict reliably is not only a clinical problem. It is a business problem that manifests on three fronts.
The first is retention. Users of preventive medicine platforms stay engaged when they perceive the platform is delivering something they couldn't get on their own: trend reading, pattern recognition, the alert at the right moment. When the alerts are inconsistent, when the presented trend doesn't make sense with what the user knows about their own health, trust erodes and engagement falls. The platform that fires false alerts from methodological artifact is training its users to ignore its alerts.
The second is institutional credibility. Preventive medicine platforms frequently sell to corporate clients, insurers, and wellness programs, which evaluate service effectiveness based on measurable outcomes. When the platform can't demonstrate, with reliable data, that its early detection effectively anticipated events and generated savings, the corporate contract becomes vulnerable at renewal. Measuring the effectiveness of the service itself depends on the same comparability that clinical prediction requires.
The third is product defensibility. A platform whose predictive capacity depends on a robust harmonization layer has a differentiator that is hard to replicate. A platform that only stores and displays tests, even with a sophisticated interface, competes with any competitor that does the same, and ultimately with generic storage solutions. The depth of the data layer is what separates a defensible product from a commoditizable one.
The data layer as infrastructure, not in-house engineering
Building internally the layer that harmonizes biomarkers from heterogeneous sources is a substantial engineering investment: parsing reports in varied formats, mapping to recognized standards like LOINC, identifying methodology and unit system, normalizing reference ranges, validating consistency, and maintaining it continuously as laboratory report formats change. For a preventive medicine healthtech, this work consumes the scarcest resource, engineering time, without generating competitive differentiation, because it is the same layer every competitor needs to build.
The literature on health interoperability recognizes that the semantic structuring of clinical data, and not just its technical integration, is what solves the problem of comparability between distinct sources.³ Standards like LOINC exist precisely for this purpose, and were developed over decades by the global health ecosystem.⁴ The healthtech that tries to rebuild this infrastructure internally is redoing work that already exists as available infrastructure.
This is where OpenHealth Technologies operates. The platform automatically correlates multiple data streams with rigorously validated logical layers of laboratory tests, receiving tests from any source and in any format, identifying methodology, unit system, and reference range, and delivering to the healthtech genuinely comparable biomarker time series, mapped to LOINC, with over 3,500 biomarkers covered. For preventive medicine platforms, this means predictive capacity, the product's central value proposition, starts operating over data in which the detected variability is the user's real variability, not the laboratory's methodological artifact. The platform stops measuring and starts predicting, which is what it always promised to do.