When the Drug Exists but the Patient Cannot Be Found
Precision oncology has entered a new phase. For many cancers, the limiting factor is no longer the availability of targeted therapies, but the healthcare system’s ability to identify eligible patients consistently and at scale.
The principle of precision medicine is straightforward: define a biological subtype with sufficient precision, develop a drug that targets it, and match the right patients to the right treatment. Yet in practice, that final step, reliably identifying the right patients, is where the system most often falls short.
This is not a problem unique to any single drug or market. It is a structural challenge that runs across precision oncology, and it becomes more acute in healthcare systems where diagnostic infrastructure is fragmented, capacity is uneven, and the patient population differs meaningfully from the cohorts that generated the underlying clinical evidence. India presents all three of these conditions simultaneously, making it one of the most important and most challenging environments in which to close the gap between clinical trial results and real-world therapeutic impact.
HER2-low breast cancer illustrates the problem with particular clarity. Drugs like Enhertu (trastuzumab deruxtecan) have demonstrated strong clinical results in this population, and recent regulatory approvals in markets including India have opened access to patients who previously had limited options. (1) But the clinical promise of these therapies can only be realized if the patients they are designed to treat are correctly identified in routine practice. In India, that identification is far less reliable than the trial data might suggest, and the reasons for that gap are structural, not incidental.
The implications extend beyond clinical practice. For pharmaceutical companies, diagnostic variability affects patient access, market penetration, and the ability to accurately estimate eligible populations. For healthcare providers, it introduces uncertainty into treatment decisions. For regulators, it raises important questions about standardization and equitable access. For investors and health system leaders, it highlights why diagnostic and data infrastructure are becoming increasingly important components of precision medicine value creation.
HER2-Low Identification: A Known Problem, an Underappreciated Scale
HER2-low is not a newly discovered biological category. It has always existed within the breast cancer landscape. What changed is that therapeutic development has now reached a point where drugs work in it, and that shift has made visible a problem the oncology world has long understood in principle but rarely addressed at scale: routine HER2 IHC scoring is inconsistent.
The difference between a HER2-zero and a HER2-low result, specifically the distinction between an IHC score of 0 and a score of 1+, is a judgment made under a microscope by a pathologist, on a given slide, on a given day, in a given laboratory. Published studies have demonstrated that the same slide can receive different scores from different pathologists, and that the same pathologist can score the same slide differently across separate reads.(2,3) This is not a reflection of poor practice. It is an inherent property of a semi-quantitative scoring system being applied to a continuous biological variable.
In high-volume, well-resourced pathology centres in Western markets, this variability is a recognized challenge that institutions manage through quality programs, digital pathology infrastructure, and second-opinion workflows. In India, the problem is structurally amplified. Pathology capacity is heavily concentrated in urban tertiary centers. Across the thousands of diagnostic laboratories serving tier-2 and tier-3 cities, there is significant variability in equipment, staining protocols, slide preparation standards, and access to subspecialty pathology expertise. The patient population presenting with breast cancer in India also differs meaningfully from Western cohorts: younger on average, with a higher proportion of aggressive subtypes and later-stage presentations at diagnosis. (4)
For any therapy whose addressable population depends on a single, difficult scoring threshold, this variability is not a minor inconvenience. It represents a systematic risk that will suppress patient identification, distort treatment allocation, and ultimately limit the real-world impact of treatments that clinical trials have shown to be effective. HER2-low is a current and visible example of this dynamic, but the underlying problem extends to any biomarker-defined subtype where diagnostic precision at the boundary is clinically consequential.
Why AI Can Help, and Why Indian Data Is the Prerequisite
The instinct within the oncology diagnostics community has been to address IHC variability through updated scoring guidelines and expanded pathologist training. Both matter, and neither should be abandoned. But guidelines require interpretation, and training requires time and sustained institutional investment. Neither scale across a market as large and structurally complex as India within a timeframe that matches the pace of therapeutic development.
One of the few approaches capable of scaling across highly distributed healthcare systems is AI-assisted pathology. Computational pathology tools have the potential to improve consistency and reproducibility in HER2 scoring while supporting pathologists within existing workflows. The clinical logic is clear. The implementation challenge is not the algorithm alone; it is the data.
AI models trained predominantly on Western pathology datasets carry embedded assumptions that are not always visible and not always benign. Tissue fixation protocols differ. Staining intensity and background characteristics vary across laboratory systems. Scanner hardware introduces its own variation in digital slide quality. And the patient population itself, the histological and receptor-status distribution of the tumours those models were trained to classify, reflects a demographic and biological reality that is meaningfully different from what presents in Indian clinical practice. (4,5)
A model trained to distinguish HER2-zero from HER2-low in slides from UK or US biobank cohorts may not generalise reliably to slides from a diagnostic laboratory in Chennai or Jaipur. This is not a theoretical concern. It is a known challenge in applied machine learning: model performance degrades when the distribution of the deployment population diverges from the training population. In the context of HER2-low scoring, where the margin between categories is already narrow and consequential, that degradation has direct clinical implications.
The prerequisite for AI-assisted HER2 standardisation that actually works in India is locally curated, locally representative pathology data: whole slide images from Indian clinical settings, annotated at score-level granularity, with sufficient volume and demographic diversity to support models that are calibrated to the population they will serve, not the population they were built on. This principle applies equally to any biomarker where AI-assisted scoring is being considered for deployment in the Indian market.
This Is Where PaiX Navigator Comes In
At Paicon, we have built PaiX Navigator as a platform for exactly this kind of work. It brings together curated, real-world IHC-annotated whole slide image data, including data sourced from Indian clinical settings, with the analytical infrastructure to explore biomarker cohorts, model patient eligibility, and support AI development workflows, all within a governed, research-ready environment.
The strength of PaiX Navigator in this context is specific and measurable. It allows researchers and pharmaceutical partners to:
-
Interrogate local cohorts: filtering by tumour type, IHC marker, score-level result, grade, specimen type, and age to understand what the real-world HER2-low population actually looks like in the Indian clinical setting
-
Model eligibility at scale: combining ER, PR, HER2, and Ki67 data simultaneously to map the full breast cancer subtype landscape across a locally representative dataset
-
Identify data gaps: understanding where scoring variability is highest, which patient subgroups are underrepresented, and where AI standardization tools would have the greatest clinical and commercial impact
-
Support AI model development: providing the annotated, structured, slide level data needed to train and validate HER2 scoring models that are calibrated to Indian pathology practice, not just global averages
For a drug like Enhertu, where the commercial frontier is moving from HER2-positive into HER2-low, and where India represents a vast, underserved, and newly accessible market, having a data partner that understands the local landscape is not a peripheral advantage. It is central to the market access strategy.
The Next Investment in Precision Oncology
The broader lesson from the HER2-low example is one that applies across precision oncology: therapeutic advancement and diagnostic infrastructure must develop in parallel. A drug that works in a precisely defined patient population only delivers its clinical benefit if that population can be reliably identified in routine practice, across the full range of healthcare settings where patients present.
That requires investing in the diagnostic and data infrastructure that makes accurate biomarker classification possible in markets like India: AI tools trained on local pathology data, validated in local laboratory settings, and deployed with the governance frameworks that clinical use demands. It means treating data infrastructure as a core component of market access strategy, not a downstream problem to be solved after approval.
The oncology industry has made enormous investments in understanding which drugs work in which precisely defined patient populations. The complementary investment, in the infrastructure that reliably identifies those populations in the real world, is where the next frontier of work lies.
Precision oncology has demonstrated what is possible when biology, therapeutics, and technology converge. The next phase of progress will depend on whether healthcare systems can build the diagnostic and data infrastructure needed to translate those advances into routine clinical practice.
For precision oncology in India, that frontier is a data problem. And data problems, approached rigorously, have data solutions.
References
-
Economic Times. AstraZeneca’s Enhertu gets regulatory approval in India for first-line HER2-positive breast cancer [Internet]. 2025 [cited 2025 Jun].
-
Denkert C, Seither F, Schneeweiss A, et al. Clinical and molecular characteristics of HER2-low-positive breast cancer: pooled analysis of individual patient data from four prospective, neoadjuvant clinical trials. Lancet Oncol. 2021;22(8):1151-1161.
-
Tarantino P, Hamilton E, Tolaney SM, et al. HER2-low breast cancer: pathological and clinical landscape. J Clin Oncol. 2020;38(17):1951-1962.
-
Agarwal G, Pradeep PV, Aggarwal V, Yip CH, Cheung PS. Spectrum of breast cancer in Asian women. World J Surg. 2007;31(5):1031-1040.
-
Tan PH, Ellis I, Allison K, et al. The 2019 World Health Organization classification of tumours of the breast. Histopathology. 2020;77(2):181-185.
