Before antibiotics, before vaccines, and before modern pediatrics, repeated febrile illnesses were simply considered part of growing up. Parents expected them. Physicians reassured families about them. Most children recovered completely and moved on.

Even today, fever remains one of the most common reasons children are brought to clinics around the world.

But what if repeated inflammatory stress during early childhood leaves subtle long-term effects on brain development?

And what if those effects only become visible decades later - not in a laboratory experiment, but through the careful analysis of large human health datasets?

That was the question behind one of our recent studies examining the relationship between repeated childhood febrile episodes and the later development of Attention Deficit Hyperactivity Disorder (ADHD).

ADHD is often discussed primarily in genetic or behavioral terms. Genetics unquestionably plays a major role. Family studies consistently show strong heritability, and many genes affecting neurodevelopment contribute to risk.

But genetics rarely acts alone.

Modern epidemiology increasingly suggests that early environmental exposures, particularly those involving inflammation, immune activation, infection, or physiologic stress during critical periods of brain development, may also influence long-term neurodevelopmental outcomes.

The challenge is that these effects are often subtle.

No single fever episode causes ADHD.

No single infection determines a child’s future.

But across hundreds of thousands of children, patterns sometimes emerge.

And that is where large-scale population data becomes scientifically powerful.

In our study, we analyzed medical records from Leumit Health Services (LHS) in Israel, examining more than 18,000 children diagnosed with ADHD and over 37,000 carefully matched controls without ADHD. The groups were matched by age, sex, socioeconomic status, ethnicity, geographic region, and duration of medical follow-up.

Rather than relying on parental memory, we examined objectively measured temperatures recorded during physician visits during the first four years of life.

This distinction matters.

Human memory is imperfect. Parents of children later diagnosed with ADHD might understandably remember childhood illnesses differently. Epidemiologic studies often struggle with this type of recall bias.

But electronic medical records allow something different:
they allow researchers to examine measurements that were recorded years before anyone knew which children would later develop ADHD.

That is one of the quiet revolutions of modern epidemiology.

When we analyzed the data, a striking pattern appeared.

Children who experienced more recorded febrile episodes during early childhood had higher rates of later ADHD diagnosis.

More importantly, the relationship followed a dose-response pattern.

The higher the fever, and the more frequently high fevers occurred, the stronger the association became.

Children with no recorded fever episodes above 37.5°C during early childhood had significantly lower ADHD rates.

Meanwhile, children with repeated high fevers - especially temperatures above 39.5°C occurring on multiple days - showed substantially higher odds of later ADHD diagnosis.

Dose-response relationships are particularly interesting in epidemiology.

When risk gradually increases alongside the intensity or frequency of an exposure, researchers pay attention, because such patterns are often indicative of a causal relationship, linking cause and effect.

Why might this occur?

There are several possibilities.

One possibility is direct inflammatory effects on the developing brain.

Early childhood is an extraordinary period of neurologic development. During the first years of life, the brain undergoes rapid synapse formation, white matter maturation, and organization of neural circuits involved in attention, executive function, and emotional regulation.

Inflammatory processes occurring repeatedly during this developmental window could theoretically influence these pathways.

Fever itself is not merely “high temperature.”

It is part of a coordinated immune response involving cytokines, prostaglandins, vascular signaling, and activation of inflammatory pathways throughout the body.

Our findings also fit within a broader and growing body of research linking infection, inflammation, and immune activation to later neuropsychiatric outcomes. Other studies, many by our group, have associated maternal fever during pregnancy, childhood infections, and immune system conditions such as G6PD deficiency and IgA deficiency with ADHD and other psychiatric disorders.

What makes this field particularly fascinating is that it sits at the intersection of neurology, immunology, developmental biology, and epidemiology.

For decades, medicine often treated the brain and immune system as largely separate domains.

That view is increasingly difficult to maintain.

Modern research increasingly shows that immune signaling, inflammation, microbiology, and neurodevelopment are deeply interconnected.

The brain does not exist in isolation from the body.

Large-scale health data help unravel just how interconnected these systems may be.

When millions of longitudinal medical records are analyzed together, hidden patterns begin to emerge across systems and across time.

Ordinary childhood illnesses may influence adult health.
Immune activation may shape neurodevelopment.
Early-life biology may leave long shadows.

And many of these relationships only become visible when researchers are willing to follow the data carefully, cautiously, and honestly - even when the findings challenge older assumptions.

That, ultimately, is one of the most important roles of modern epidemiology: to help us notice the patterns that human intuition alone would never see.

References:

1. The Association Between Repeated Measured Febrile Episodes During Early Childhood and Attention Deficit Hyperactivity Disorder: A Large-Scale Population-Based Study.

   Israel A, Merzon E, Krone B, Faraone SV, Green I, Golan Cohen A, Vinker S, Cohen S, Ashkenazi S, Magen E, Weizman A, Manor I.
   The Journal of Attention Disorders 2024 Mar;28(5):677-685.

   doi:10.1177/10870547231215289.

2. Increased Prevalence of Attention Deficit Hyperactivity Disorder in Individuals with Selective Immunoglobulin A Deficiency: A Nationwide Case-Control Study.

   Merzon E, Farag R, Ashkenazi S, Magen E, Manor I, Weizman A, Green I, Golan-Cohen A, Genshin A, Vinker S, Israel A.
   Journal of Clinical Medicine 2024 Oct 12;13(20):6075.
   doi:10.3390/jcm13206075.

Modern medicine provides highly effective medications for relieving these symptoms: proton pump inhibitors (PPIs), which are medications that suppress gastric acid secretion (omeprazole, lansoprazole, and esomeprazole...).

Over the past decade, however, a growing number of observational studies have reported that upper gastrointestinal cancers, including stomach and esophageal cancers, are substantially more common among long-term PPI users.

Such a risk is biologically plausible and should not be dismissed lightly. Chronic acid suppression can alter stomach physiology, increase gastrin levels, and change the gastric microbiome in ways that could potentially increase cancer risk.

This creates a real dilemma for treating physicians. Should they prescribe medications that effectively relieve patients’ symptoms, or avoid a treatment that could contribute to future cancer risk?

To answer this question properly, it is crucial to separate the effects associated with PPI use from the symptoms and underlying conditions for which these medications are prescribed.

Upper gastrointestinal cancers often begin with vague symptoms such as reflux, gastritis, abdominal pain, nausea, and indigestion, precisely the symptoms for which PPIs are prescribed. This creates a major epidemiologic challenge known as reverse causation, where medications may appear associated with cancer, not because they cause cancer, but because underlying pathologic processes caused the symptoms that led to the treatment.

Notably, patients receiving PPIs have higher prevalence of gastrointestinal disease, smoking, alcohol exposure, obesity, and Helicobacter pylori infection, all of which may independently contribute to cancer risk.

So how can we separate the effect of the medication from the effect of the symptoms and diseases that caused patients to take the medications?

To address this question, we analyzed longitudinal electronic health records from our health system, identifying 875 patients diagnosed with upper gastrointestinal cancers during the last two decades and matching them to 8,750 similar control individuals who had not been diagnosed with cancer.

When we initially examined PPI exposure through pharmacy records during the 5 years preceding diagnosis, we were able to reproduce the concerning associations reported by earlier studies. Patients exposed to the major classes of PPIs had substantially higher rates of upper gastrointestinal cancer, with stronger associations observed for medication with more effective acid suppression: approximately a 2.4-fold increase for omeprazole and a 4-fold increase for esomeprazole.

But then we performed a more refined analysis.

Instead of analyzing the entire 5-year exposure period as a single variable, we divided the years preceding diagnosis into multiple smaller intervals and incorporated them simultaneously into a multivariable regression model.

This approach allows to estimate the independent contribution of medication exposure during each specific time window

The results revealed a striking temporal pattern.

The strongest apparent associations between PPIs and cancer occurred immediately before diagnosis, precisely during the period when early cancers are most likely to produce severe symptoms for which patients take treatment.

For example:

• esomeprazole exposure during the final months before diagnosis was associated with a 7.7-fold increase in risk,

• omeprazole exposure with a 3.8-fold increase,

• lansoprazole exposure with a 3.5-fold increase.

However, when exposure occurred further from diagnosis, the signal changed dramatically. Exposure occurring 1-2 years, 2-3 years, and especially 3-10 years before diagnosis shifted toward neutral or even reduced risk. If PPIs were truly driving cancer development, one would expect longer and earlier exposure to be associated with progressively higher risk.

The pattern became even clearer after integrating gastrointestinal diagnoses into the analysis.

The strongest associations with future cancer risk were not linked to the medications themselves, but to the clinical conditions for which the medications were prescribed: peptic ulcer disease, abdominal pain, Gastro-esophageal reflux disease (GERD), gastritis, alcohol use, and Helicobacter pylori infection.

Moreover, once we excluded the periods immediately preceding diagnosis (where medication exposure is most likely to reflect new onset of symptoms), PPI exposure was no longer associated with increased cancer risk, and was instead associated with reduced risk.

These findings suggest that much of the alarming associations reported in earlier observational studies reflected symptoms experienced by patients who took these medications rather than harmful effects of the medications themselves. In fact, by reducing excess acidity and chronic inflammation, these medications may even reduce long-term cancer risk by limiting chronic tissue damage.

The broader lesson extends beyond PPIs. Large healthcare databases are powerful scientific tools, but interpreting them correctly requires careful attention to timing, disease progression, clinical context, and epidemiologic bias.

Using appropriate methodology, our study provides reassuring evidence that appropriately used PPIs are unlikely to be major long-term drivers of upper gastrointestinal cancer risk.

At the same time, the findings contain an important clinical warning: physicians should not ignore persistent or newly appearing upper gastrointestinal symptoms simply because acid suppression relieves them.

Sometimes the medication is not the danger.

Sometimes the symptoms themselves are a signal that should not be ignored.

References:

1. Association between proton pump inhibitor use and upper gastrointestinal cancer: A matched case-control study accounting for reverse causation and confounding by indication.

   Sawaid IO, Din Z, Golan E, Ruppin E, Golan-Cohen A, Green I, Merzon E, Vinker S, Samson AO, Israel A.
   PLoS Medicine 2026;23(1).
   doi:10.1371/journal.pmed.1004842

2. Context matters: Causality and global epidemiology of proton pump inhibitor safety.

   Okabayashi S, Kaplan GG.

   PLoS Medicine 2026;23(1).
   doi:10.1371/journal.pmed.1004862
   

After an initial period of lockdowns and overflowing hospitals, the vaccination campaign quickly reduced the rate of new COVID-19 cases and severe disease across the country. Israel quickly became a global model for large-scale vaccination. Restrictions were lifted. Daily life began returning to normal. There was genuine optimism that the pandemic might be ending.

At the Leumit Research Institute, where we continuously analyze nationwide longitudinal electronic health records, we began tracking the immune response to SARS-CoV-2 in the population.

The early antibody response following mRNA vaccination was remarkably strong.

In individuals vaccinated with two doses of the Pfizer-BioNTech vaccine, antibody titers often reached levels around 10,000 AU/mL shortly after vaccination, substantially higher than the antibody levels typically observed after natural SARS-CoV-2 infection, which were usually closer to 1,000 AU/mL.

But when we began examining how antibody levels changed over time, an unexpected pattern emerged.

The immune response induced by vaccination was not merely declining. It was declining very rapidly.

Using large-scale longitudinal serologic data from thousands of individuals across Israel, we observed that antibody titers following vaccination decreased by approximately 40% every month. In contrast, antibody levels following natural infection declined much more slowly, by only around 4% per month.

The difference was striking.

Although antibody levels following vaccination were initially higher than those observed after natural infection, the decline was so steep that within a few months they fell below those observed following infection.

The graphs below from our original publication illustrate this phenomenon.

Following vaccination (top), antibody titers began at very high levels but rapidly declined over time. Following natural infection (bottom), antibody levels were lower initially, but remained comparatively stable over many months. Importantly, the vertical scale in these graphs is logarithmic, meaning that each vertical step represents a tenfold decrease in antibody levels.

At the time, these findings were uncomfortable.

The immunologic dynamics suggested that although the vaccines initially appeared highly effective at reducing severe disease, the protection they conferred against infection might not remain stable for long.

Based on the observed rate of antibody decay, we became concerned that within several months, vaccine-induced protection against infection would substantially weaken, leaving vaccinated individuals increasingly susceptible to breakthrough infections.

And indeed, several months later, breakthrough infections began rising sharply in Israel during the emergence of the so-called Delta wave in the summer of 2021.

At the time, the dominant explanation was that the Delta variant had evolved mutations that enabled it to evade vaccine-induced immunity, raising concerns that protection against infection might require updated vaccines specifically adapted to the new variant.

But when we examined nationwide health data, another signal clearly emerged: the risk of infection was strongly associated with the time elapsed since vaccination.

Individuals vaccinated earlier were substantially more likely to develop breakthrough infections.

We analyzed RT-PCR results from more than 83,000 adults who had received two vaccine doses. After adjusting for demographic factors and comorbidities, the risk of infection increased progressively with time since vaccination. Individuals vaccinated more than six months earlier had nearly a threefold increased risk of infection compared with recently vaccinated individuals.

Importantly, this pattern was observed across all age groups.

At the time, this remained a controversial conclusion.

The vaccines had initially been viewed as providing relatively durable protection against infection, and the idea that immunity might substantially wane within months was uncomfortable both scientifically and politically.

Nevertheless, the data were difficult to ignore.

We rapidly disseminated these findings as a preprint, and the evidence regarding waning immunity contributed to the decision by the Israeli COVID-19 response task force to launch a nationwide booster vaccination campaign, initially targeting elderly and vulnerable individuals.

The effect of this decision was substantial.

Individuals who received the booster dose experienced a marked reduction in breakthrough infections and severe disease. The booster campaign was subsequently expanded to the general population in Israel, and similar policies were later adopted internationally.

Our study was later published in the British Medical Journal (BMJ).

This experience taught an important lesson about science and public health.

Scientific evidence is not static, and science should never become dogmatic.

While early clinical trials provide essential answers, large-scale real-world epidemiology remains indispensable. Population-level health data have the unique ability to reveal biological dynamics that are difficult to detect in controlled trials, especially when trial reports are based on limited follow-up.

We pragmatically followed vaccination effects at the population level and concluded that although mRNA vaccines substantially reduced severe disease and mortality among elderly and vulnerable individuals, they also had important limitations, including rapidly waning protection against infection among vaccinated individuals.

Recognizing these limitations early was essential for adapting public-health policy to a changing reality.

Good science requires the willingness to follow the data without preconceived conclusions, and to report findings honestly, even when the signal is uncomfortable.


References

1. Large-Scale Study of Antibody Titer Decay following BNT162b2 mRNA Vaccine or SARS-CoV-2 Infection.
   Israel A, Shenhar Y, Green I, Merzon E, Golan-Cohen A, Schäffer AA, Ruppin E, Vinker S, Magen E.
   Vaccines (Basel). 2021 Dec 31;10(1):64.
   doi:10.3390/vaccines10010064.

2. Elapsed time since BNT162b2 vaccine and risk of SARS-CoV-2 infection in a large cohort. medRxiv. Posted August 5, 2021.
   Israel A, Merzon E, Schäffer AA, Shenhar Y, Green I, Golan-Cohen A, Ruppin E, Magen E, Vinker S.
   doi:https://doi.org/10.1101/2021.08.03.21261496

3. Elapsed time since BNT162b2 vaccine and risk of SARS-CoV-2 infection: test negative design study.
   Israel A, Merzon E, Schäffer AA, Shenhar Y, Green I, Golan-Cohen A, Ruppin E, Magen E, Vinker S.
   BMJ. 2021 Nov 24;375.
   doi:10.1136/bmj-2021-067873.

What if one of the most widely used tests in modern medicine systematically underestimates diabetes severity in millions of people?

Over the past two decades, Hemoglobin A1c (HbA1c) testing has become central to diabetes diagnosis and management. HbA1c measures the percentage of hemoglobin molecules inside red blood cells that become coated with glucose over time. Since red blood cells normally circulate in the bloodstream for approximately three months, HbA1c provides physicians with an estimate of the average blood glucose level over the preceding weeks.

Doctors use HbA1c constantly:

• to diagnose diabetes,

• to estimate disease severity,

• to decide whether treatment should be intensified,

• to monitor long-term glucose control,

• and in many healthcare systems, to determine eligibility for advanced medications.

But HbA1c depends on something many people do not realize:
the lifespan of red blood cells.

If red blood cells break down earlier than usual, they have less time to accumulate glucose. HbA1c values may then appear artificially low, even when glucose levels are dangerously elevated.

One of the most common conditions affecting red blood cell survival is Glucose-6-Phosphate Dehydrogenase (G6PD) deficiency.

G6PD deficiency is the most common enzymatic deficiency in humans, affecting more than 400 million people worldwide. It is particularly common among individuals of African, Mediterranean, Middle Eastern, and parts of Asian ancestry. In the United States alone, approximately 10% to 14% of Black American men carry G6PD deficiency.

Yet despite its enormous prevalence, G6PD deficiency remains largely overlooked in diabetes care.

In our recent studies using nationwide longitudinal electronic health records from Israel, we examined more than two decades of data comparing thousands of individuals with G6PD deficiency to matched controls.

The findings were striking.

At the same glucose levels, HbA1c values were substantially lower in patients with G6PD deficiency. A HbA1c value of 6.5%, the standard diagnostic threshold for diabetes, corresponded to fasting glucose levels around 168 mg/dL in men with G6PD deficiency, compared with 126 mg/dL in the general population.

This means that many patients with significant hyperglycemia may appear substantially less diabetic than they truly are.

The scatter plots below illustrate this phenomenon.

Each point represents paired glucose and HbA1c measurements obtained from patients. The blue regression line represents individuals without G6PD deficiency, while the orange line represents individuals with G6PD deficiency. At every glucose level, the HbA1c values in G6PD-deficient individuals are shifted downward.

The discrepancy becomes even larger at higher glucose levels.

For example, among individuals without G6PD deficiency, a HbA1c value of 7.5% corresponds approximately to glucose levels around 165 mg/dL. In contrast, in men with G6PD deficiency, the same HbA1c value corresponds to glucose levels above 210 mg/dL.

The larger effect observed in men is not incidental.

The G6PD enzyme is encoded on the X chromosome. Because males carry only one X chromosome, a defective G6PD gene usually affects nearly all of their red blood cells, often producing more pronounced biological effects.

Females, however, carry two X chromosomes. Early during embryonic development, each stem cell randomly “switches off” one of the two X chromosomes. As a result, many women become biological mosaics: some of their red blood cells express the normal G6PD gene, while others express the deficient version. This partial “rescue” often reduces the severity of the condition in females, although the degree of protection can vary substantially between individuals depending on the proportion of cells expressing each X chromosome.

This has enormous clinical implications.

Modern diabetes care increasingly relies on powerful medications such as GLP-1 receptor agonists and SGLT2 inhibitors. These medications not only lower glucose, but also reduce cardiovascular disease, kidney failure, hospitalization, and mortality.

However, in many healthcare systems, access to these medications depends heavily on HbA1c thresholds.

If HbA1c systematically underestimates disease severity, patients may receive treatment later, less aggressively, or not at all.

That is exactly what we observed.

To study this, we used a statistical approach called survival analysis.

Despite its name, survival analysis is not limited to mortality studies. It is a method used to analyze how long it takes until a particular event occurs. In our case, the “event” was not death, but the initiation of diabetes treatment or the development of severe diabetic complications.

The Kaplan-Meier curves below show the cumulative probability over time of patients receiving diabetes medications after meeting objective criteria for diabetes based on glucose levels.

The curves clearly demonstrate that patients with G6PD deficiency experienced substantial delays in treatment initiation after their fasting blood test first reached daiebetes levels.

More importantly, they were significantly less likely to receive the newer classes of diabetes medications - particularly GLP-1 receptor agonists (such as Victoza, Trulicity, and Ozempic) and SGLT2 inhibitors (such as Jardiance and Forxiga) - therapies that have revolutionized our ability to reduce diabetes complications, kidney failure, cardiovascular disease, and mortality.

At the same time, these patients experienced markedly higher rates of severe diabetic complications, including kidney disease, cardiovascular complications (myocardial infarction, stroke), and diabetic neuropathy.

Paradoxically, reliance on HbA1c may not only contribute to undertreatment, but potentially worsen the biological dynamics of diabetes itself.

Some older diabetes medications, particularly sulfonylureas, may increase oxidative stress within red blood cells. In patients with G6PD deficiency, whose red blood cells are already vulnerable to oxidative damage, this may accelerate hemolysis - the breakdown of red blood cells.

As red blood cells are destroyed more rapidly, HbA1c values may fall even further, creating the illusion of improving diabetes control despite worsening hyperglycemia.

In other words:
the biomarker itself may become progressively misleading precisely in the patients who are at highest risk.

Importantly, this problem is not limited to one ethnic group or one country.

Similar effects of G6PD deficiency on HbA1c levels have now been reported in African-American populations and East Asian populations carrying distinct G6PD genetic variants.

This suggests that a substantial fraction of diabetes inequity observed worldwide may partly reflect hidden biological bias embedded within one of medicine’s most widely used laboratory tests.

The encouraging aspect is that this inequity may be remediable.

Awareness matters.

Clinicians caring for patients from high-prevalence populations should recognize that HbA1c may substantially underestimate glycemic burden in G6PD-deficient individuals. Alternative approaches, including fasting glucose measurements, continuous glucose monitoring, fructosamine, glycated albumin, or modified interpretation of HbA1c values, may improve care in these populations.

Large-scale health data sometimes reveals hidden biological patterns that remain invisible at the level of individual patients.

This is one of them.

And it may help explain why some populations continue to experience disproportionate diabetes complications despite decades of medical progress.

References:
Type 2 Diabetes in Patients with G6PD Deficiency.
The New England journal of medicine. 2024 Aug 8; 391(6):568-569.
Israel A, Raz I, Vinker S, Magen E, Green I, Golan-Cohen A, Berkovitch M, Merzon E.
doi: 10.1056/NEJMc2406156.

More on Type 2 Diabetes in Patients with G6PD Deficiency. Reply.
The New England journal of medicine. 2024 Oct 31; 391(17):1664.
Israel A, Vinker S, Merzon E.
doi: 10.1056/NEJMc2411261#sa2.

Health disparities in diabetes treatment: The challenge of G6PD deficiency.
Diabetes research and clinical practice. 2025 Jan; 219:111965.
Israel A, Raz I, Green I, Golan-Cohen A, Berkovitch M, Magen E, Vinker S, Merzon E.
doi: 10.1016/j.diabres.2024.111965.