In the summer of 1948, a heat wave swept New York City, and paediatrician Paul di Sant'Agnese noticed something odd among the children being treated for heat stroke associated with salt depletion: several had the condition known as “cystic fibrosis of the pancreas” or simply CF today. Their skin left a distinctly salty residue. Curious, di Sant'Agnese investigated further, and within a few years had shown that children with CF lose abnormally large amounts of salt through their sweat. In 1959, Gibson and Cooke described the pilocarpine-iontophoresis method for measuring sweat sodium and chloride, which became the basis of the modern sweat test and still remains the diagnostic test for CF today. It is a nice reminder that some of medicine's most durable diagnostic tools began as a clinician noticing something that didn't quite fit.
What followed over the next seventy-odd years is one of the more remarkable trajectories in clinical medicine. From the 1950s and 1960s, specialised, multidisciplinary CF centres — bringing respiratory physicians, dietitians, physiotherapists, gastroenterologists and endocrinologists into one coordinated team — began to emerge and progressively improve outcomes. When the Cystic Fibrosis Foundation was established in 1955, children with CF rarely lived long enough to attend elementary school; over the following decades, specialised care, antibiotics, nutritional therapy and airway-clearance strategies transformed survival. Pancreatic enzyme replacement therapy transformed nutritional management, helping address the severe malabsorption, poor growth and malnutrition that had characterised pancreatic-insufficient CF. In 1989, an international collaboration identified the CFTR gene — the gene responsible for cystic fibrosis — a landmark that took a further 23 years to translate into ivacaftor, the first drug to directly correct the CFTR protein defect, in patients with an amenable mutation. The arrival of broader triple-combination modulator therapy in 2019–2020 has transformed CF care again. Contemporary CF Foundation registry data now project median survival of around 66 years for children with CF born in 2021–2025.
Problems that only appear once people live long enough to have them
One consequence of that lengthening lifespan is that CF medicine has had to keep discovering ways of dealing with new problems that simply didn't occur when life expectancy was shorter. Cystic fibrosis-related diabetes (CFRD) is a good example: as children with CF began living into adolescence and adulthood, gradual, insulin-deficiency-driven dysglycaemia started to emerge as a recognised complication, understood to erode nutritional status and lung function in ways not captured by the traditional two-point glucose tolerance test. Our own contribution to that story — describing which glucose thresholds actually predicted decline (Diabetes Care, 2010), piloting a simpler once-daily insulin regimen to reduce treatment burden (Archives of Disease in Childhood, 2012), and ultimately testing early insulin treatment in the multicentre CF-IDEA randomised trial across Australia and the United States (The Lancet Child & Adolescent Health, 2025) — sits as one small part of that decades-long pattern of CF medicine noticing a new problem as patients lived long enough to develop it, and building the evidence to treat it well.
That pattern hasn't stopped. With modulator therapy now extending life expectancy further still, a new question is emerging: will people with CF, living decades longer and often gaining weight for the first time in the disease's history, begin to develop obesity, dyslipidaemia, insulin resistance and other metabolic complications that were previously uncommon in CF — and will they eventually experience the cardiovascular and other long-term complications that have historically been rare in a population that seldom survived long enough for them to emerge? Colorectal cancer is another emerging concern in the ageing CF population, and here too the gut may be offering an early warning: work led by Professor Chee (Keith) Ooi and colleagues at UNSW and Sydney Children's Hospital has found that a genotoxin-producing strain of E. coli in the gut microbiome is strongly associated with ileocolonic neoplasia in adults with CF, raising the possibility that this could eventually become a microbial biomarker of colorectal cancer risk — if the finding is validated in larger cohorts (van Dorst et al., Journal of Cystic Fibrosis, 2026).
Where AI could plausibly compress the next seventy years into a much shorter one
The CF story took seven decades from a clinician's bedside observation to a therapy that corrects the molecular defect, and even the fastest-moving part of it — gene discovery to modulator — took over two decades. A few places where current AI tools could plausibly shorten the equivalent arc for the next generation of CF questions:
- Clinical observation at scale. The original CF story began with a clinician noticing that something did not fit the expected pattern. Modern healthcare generates millions of similarly fragmented observations every day, scattered across clinical notes, pathology, imaging, medications, registries and correspondence. AI could help detect recurring patterns across this information, turn them into testable hypotheses, and connect clinicians encountering the same signal — potentially shortening the path from bedside observation to collaborative research.
- Structure-based drug design. AI-based protein-structure prediction and molecular-design tools can help identify and prioritise candidate molecules before laboratory synthesis and testing — relevant to the ongoing search for modulators covering the subset of CFTR mutations not well served by current therapies.
- Registry-scale signal detection. AI-assisted analysis of large CF registries could potentially identify emerging metabolic or other clinical signals before they become obvious in conventional hypothesis-driven studies.
- Federated multi-site collaboration. Rare disease research is often bottlenecked by assembling enough patients across enough hospitals. Federated learning allows institutions to train a shared model while keeping patient-level data within each institution, which could reduce the need to centralise sensitive datasets and make some forms of multi-site research easier to coordinate — although ethics, governance, security and institutional approvals remain essential regardless. AI may not eliminate research governance; it may simply change what needs to be governed.
- Automated evidence synthesis. The post-modulator metabolic literature is already accumulating quickly; AI-assisted systematic review tools can help guideline and consensus documents keep pace, rather than lagging years behind as literature synthesis traditionally has.
- Predictive risk stratification. If validated prospectively, models combining genotype, modulator exposure, growth trajectory and metabolic data could help identify individuals at higher risk of complications, enabling targeted surveillance rather than one-size-fits-all screening built for an earlier version of the disease.
- Drug-repurposing screens. AI-driven drug-repurposing platforms can systematically search for connections between agents already established for insulin resistance and metabolic syndrome in the general population and CF's emerging metabolic phenotype, at a scale that would be difficult to achieve manually.
It seems likely that clinicians astutely observing unusual patterns will still trigger leaps forward in clinical research, translating over time to improvements in clinical practice. The arc of improvements in CF care sparked by di Sant'Agnese's observation in a New York heatwave is a testament to the human spirit. What AI may change is not the source of medical discovery, but the speed at which observations become hypotheses, hypotheses become collaborations, and collaborations become evidence.
Selected sources: Hameed S et al., Diabetes Care 2010;33:221–226. Hameed S et al., Archives of Disease in Childhood 2012;97:464–467. Hameed S et al. (CF-IDEA), The Lancet Child & Adolescent Health 2025;9:371–382. van Dorst J, Taylor N, Pushpakumara BLDU, et al., Ooi CY. Genotoxic pks+ E. coli is strongly associated with ileocolonic neoplasia in adults with Cystic Fibrosis. Journal of Cystic Fibrosis 2026. doi.org/10.1016/j.jcf.2026.05.016.