A single CDMA dose at 4 W/kg, fewer animals, no GSM arm, and no dose–response analysis cannot overturn a larger multi-dose carcinogenicity study—especially when Korea still found NTP’s signature heart tumor only in RF-exposed rats.

Suggested URL slug: japan-korea-rf-rat-study-does-not-rebut-ntp

Meta description: The 2026 Japan–Korea RF studies tested one CDMA dose with limited statistical power. Here is why they cannot rebut the NTP cancer findings.

---

Science does not erase a dose–response curve with one dot.

Yet that is essentially how the 2026 Japan–Korea radiofrequency-radiation studies have been presented to the public. ARPANSA went so far as to headline its review: “Korean and Japanese studies agree that radiofrequency exposure does not cause cancer in rats.” That sweeping claim goes far beyond what the experiments were capable of establishing. The studies examined one modulation, at one exposure level, in male rats only, with too few animals to reliably detect the rare tumors at issue.

On July 29, 2026, Ronald L. Melnick, Joel M. Moskowitz, and Paul Héroux published a direct methodological response in Environmental Health. Their commentary explains why the Japan–Korea studies could not reasonably validate—or invalidate—the National Toxicology Program findings. This is especially significant because Melnick led the design of the original NTP radiofrequency studies. The new commentary concludes that the follow-up studies’ reduced group size, single-dose design, and omission of the NTP’s highest CDMA dose deprived them of the sensitivity required to challenge the original findings.

The bottom line is straightforward:

The Japan–Korea studies did not reproduce the NTP experiment. They sampled one narrow coordinate within the much larger exposure-response landscape that NTP investigated.

They therefore cannot be used as evidence that the NTP cancer findings were disproven.

---

The entire dispute in one table

The NTP exposed rats prenatally, during lactation, and for two years using both 900 MHz GSM and CDMA at three exposed dose levels. Its final assessment found clear evidence of malignant heart schwannomas in exposed male rats and some evidence of malignant brain gliomas. The Japan–Korea experiments used 900 MHz CDMA at only 4 W/kg, with 70 males in each cage-control, sham, and exposure group.

The Korean paper itself says the project was “not designed as a complete replication” of NTP. The Japanese authors went further, stating that their primary purpose was “not to replicate the NTP findings quantitatively.” Those acknowledgments alone should end any attempt to market these papers as a definitive rebuttal.

---

One exposure level cannot erase a dose–response relationship

The Japan–Korea experiments had a sham condition and one RF condition:

$$0\text{W/kg}\text{versus}4\text{W/kg}.$$

That is a comparison between two points. It is not a dose–response experiment.

The NTP design contained four points:

$$0,1.5,3,6\text{W/kg}.$$

Those multiple groups allowed NTP investigators to determine whether tumor incidence changed with dose, perform formal trend tests, identify potentially nonlinear patterns, and distinguish isolated findings from a response distributed across several exposure levels.

That distinction is fundamental in experimental carcinogenesis. Melnick, Moskowitz, and Héroux note that FDA and OECD carcinogenicity guidance uses at least three exposed dose levels plus a concurrent control so that a trend can be evaluated. A one-dose study eliminates trend analysis before the experiment begins.

Good Laboratory Practice cannot repair this limitation.

GLP can help ensure that animals were handled consistently, exposures were characterized properly, records were preserved, and the experiment was executed according to protocol. It cannot transform an underpowered, single-dose protocol into a sensitive multi-dose carcinogenicity test.

GLP can make a narrow answer reliable. It cannot make the answer broader than the question that was asked.

---

The follow-up design missed both forms of NTP evidence

Precision matters here because NTP did not produce only one type of response.

The most important CDMA heart finding followed a positive exposure trend. The clearest nonmonotonic brain pattern appeared in the GSM arm.

The Japan–Korea design missed both.

1. It omitted the 6 W/kg CDMA group where NTP’s heart result was strongest

In male rats exposed to CDMA-modulated RF, NTP reported malignant heart schwannomas as follows:

The mortality-adjusted trend was statistically significant at P = 0.011, and the comparison between the 6 W/kg group and sham controls was significant at P = 0.030. NTP also found Schwann-cell hyperplasia—an associated non-neoplastic lesion—in three males at 6 W/kg and none in the other CDMA groups.

The Japan–Korea project did not include 6 W/kg.

It chose 4 W/kg—between NTP’s 3 and 6 W/kg groups—and then attempted to speak to the reproducibility of the NTP result without testing the exposure at which NTP’s decisive pairwise heart comparison occurred.

That is not a small technical difference. It removes the strongest CDMA anchor from the attempted validation.

2. It omitted GSM—the modulation with NTP’s clearest nonmonotonic brain pattern

In NTP’s male GSM groups, malignant brain gliomas were:

Neither malignant glioma nor glial-cell hyperplasia occurred in the male sham group, whereas both appeared across the exposed groups. The individual comparisons did not reach conventional statistical significance, but NTP’s expert review ultimately classified the glioma evidence as some evidence of carcinogenic activity.

This is an explicitly nonmonotonic pattern: the largest raw glioma count occurred at the two lower exposure levels, not the highest.

The Japan–Korea studies did not test GSM at any dose.

A CDMA-only experiment therefore cannot be cited as a rebuttal to a GSM-associated outcome. Modulation is part of the tested exposure—not a cosmetic label that can be exchanged after the fact.

A study that never exposed a single animal to GSM cannot disprove an observed GSM response.

---

The Korean study was not tumor-free

The phrase “no statistically significant increase” is repeatedly transformed in public communications into “no cancer” or “RF does not cause cancer.”

Those statements are not equivalent.

In the Korean study, endocardial heart schwannomas—the signature rare tumor behind NTP’s strongest conclusion—were observed as follows:

The Korean authors explicitly acknowledged that the schwannomas occurred only in the RF-exposed group. The incidence was 2.9%.

That result does not, by itself, prove that RF caused those two tumors. The count is small, and rare tumors can occur sporadically.

But it is equally false to describe the result as though the exposed animals had no relevant tumors.

The direction of the result is not opposite to NTP. It is directionally compatible with it:

• zero in the concurrent cage controls;
• zero in the concurrent sham controls;
• two in the RF group.

Melnick and colleagues point out that the Korean 2.9% rate is essentially the same unadjusted incidence that NTP observed at 3 W/kg CDMA: 3 tumors among 90 males, or approximately 3.3%.

The scientifically defensible characterization is:

The Korean result was directionally consistent with NTP but statistically inconclusive because the experiment lacked sufficient power.

That is very different from saying the study proved RF was non-carcinogenic.

---

The studies were dramatically underpowered for the tumor they were supposed to verify

Rare-tumor studies require large groups because an actual increase may involve only a handful of animals.

Melnick, Moskowitz, and Héroux estimated that if the expected heart-schwannoma incidence at 4 W/kg were approximately 4.4%, a study would require:

• 258 animals per group for 80% statistical power;
• 323 animals per group for 90% power.

Korea used 70 animals per group.

Japan began with 70, but only 68 RF-exposed males were included in its final tumor table after two animals died following an equipment malfunction. Even combining the Korean and Japanese groups would yield only about 140 exposed animals—still substantially below the estimated requirement.

This is not a semantic dispute about statistical preferences. It goes directly to the conclusion the studies can support.

A study with inadequate power can easily produce:

$$P>0.05$$

even when the underlying effect is real.

That is why “not statistically significant” does not mean “demonstrated absence.” It means the study did not establish a difference at the selected statistical threshold under the design and sample size used.

When Korea observed the target rare tumor in 2 of 70 exposed rats and none of 140 concurrent control animals, the low-powered design produced exactly the ambiguity one would expect:

• too few cases to establish causation;
• too few animals to reliably exclude an effect;
• no additional dose groups to reveal a trend.

The experiment was structurally incapable of resolving the question.

---

Survival differences do not erase NTP’s mortality-adjusted result

One common argument against NTP is that exposed male rats survived longer than sham controls, potentially giving them more time to develop late-life tumors.

That issue deserved statistical treatment—and NTP treated it statistically.

NTP used the Rao–Scott-adjusted Poly-3 test, which adjusts tumor incidence for animals that die before terminal sacrifice and accounts for within-litter correlation. After mortality adjustment, the CDMA heart-schwannoma trend remained significant at P = 0.011, and the 6 W/kg pairwise comparison remained significant at P = 0.030.

The Japanese follow-up also had a large survival difference. At 105 weeks, survival was:

• 34.3% in cage controls;
• 42.9% in sham controls;
• 64.7% in RF-exposed rats.

The Japanese authors acknowledged that differential survival complicated interpretation and that their smaller single-dose design had less power to identify rare tumors.

Survival is therefore not a one-directional argument that automatically invalidates NTP. It is a time-at-risk variable requiring mortality-adjusted analysis, adequate sample size, and careful interpretation.

It cannot justify discarding NTP’s adjusted statistics while treating the much smaller follow-up studies as definitive.

---

The 4 W/kg dose was selected from the guideline being defended

The choice of 4 W/kg was not derived from a full mapping of the NTP cancer response.

The Korean paper states that 4 W/kg was chosen because it is a key reference point in international human RF-exposure guidelines. The project’s 2022 design paper likewise explained that 4 W/kg corresponds to the operational threshold from which ICNIRP’s thermal exposure limits are derived.

That history matters.

The 4 W/kg reference originated in short-term experiments examining behavioral disruption and temperature-related effects in small groups of food-deprived rats and monkeys. It was not derived from a lifetime carcinogenicity study, a developmental exposure study, or an analysis of low-dose nonlinear responses. ICNIRP and the FCC later applied reduction factors to that acute threshold to establish whole-body limits of 0.4 W/kg for occupational exposure and 0.08 W/kg for the public.

This creates an obvious risk of institutional circularity:

1. A thermal guideline identifies 4 W/kg as the key reference point.
2. A chronic study selects only that guideline-centered point.
3. The study lacks enough doses and animals to detect the original response reliably.
4. The resulting nonsignificant result is then used to defend the thermal guideline.

That is not biology-centered dose mapping. It is guideline-centered testing.

Melnick and colleagues correctly argue that an acute thermal reference point cannot establish the proper challenge dose—or the proper number of doses—for a chronic carcinogenicity validation study.

---

The RF Safe boundary-layer interpretation

At RF Safe, we describe the Japan–Korea result as a boundary-layer null.

A boundary-layer null is a negative or nonsignificant result obtained at one narrow exposure coordinate near a transition between competing biological processes. The measured endpoint reflects the net result of multiple systems:

$$\text{Observed outcome}=f(\text{carrier},\text{modulation},\text{SAR},\text{timing},\text{tissue state},\text{injury},\text{repair},\text{adaptation},\text{survival}).$$

At a given exposure, RF-induced perturbation may coexist with thermoregulation, antioxidant responses, heat-shock signaling, DNA repair, apoptosis, immune surveillance, altered metabolism, or changes in survival. The net tumor count at one dose cannot reveal how these processes vary above or below that point.

The supplied RF Safe analysis correctly identifies the central inference problem: a single 4 W/kg CDMA condition cannot falsify lower-dose, modulation-specific, or nonmonotonic biological responses.

To be precise, the available data do not prove that a particular protective response suppressed tumors at 4 W/kg. That remains a mechanistic hypothesis.

What the data prove is that the Japan–Korea design could not test that hypothesis.

A single exposed group cannot determine whether 4 W/kg lies:

• below a response peak;
• at the peak;
• above the peak;
• on a plateau;
• or within a region where injury and adaptation partially cancel one another.

The NTP GSM brain findings make this concern concrete. Malignant gliomas numbered 3, 3, and 2 across 1.5, 3, and 6 W/kg. Glial hyperplasia numbered 2, 3, and 1. Sampling only one CDMA point at 4 W/kg tells us nothing about that GSM response surface.

The proper conclusion is not that the boundary-layer hypothesis has been proven.

It is that the hypothesis remains entirely untested by the studies being advertised as a rebuttal.

---

Modulation cannot be treated as irrelevant

The follow-up researchers selected CDMA partly because it was a commonly used waveform in Korea and Japan and because NTP’s strongest statistically significant heart result appeared in the CDMA arm. That is a reasonable basis for conducting a CDMA study. It is not a basis for generalizing the result to GSM, Wi-Fi, LTE, 5G, or “radiofrequency exposure” as one undifferentiated category.

The NTP report itself found malignant heart schwannomas under both GSM and CDMA and brain gliomas associated with both modulations. The exact incidence patterns differed:

• CDMA heart schwannomas rose from 0 to 2 to 3 to 6;
• GSM heart schwannomas were 0, 2, 1, and 5;
• GSM brain gliomas were 0, 3, 3, and 2;
• CDMA brain gliomas were 0, 0, 0, and 3.

Those differences show why carrier and modulation must be preserved as experimental variables rather than collapsed into average SAR alone.

A valid test of modulation dependence requires parallel exposure arms under matched dosimetry and temperature conditions.

The Japan–Korea papers did not perform that experiment.

---

A statistically clean null was impossible because no trend could be calculated

The most important statistical information in a carcinogenicity bioassay frequently comes not from a single pairwise comparison but from the pattern across several doses.

With groups at 0, 1.5, 3, and 6 W/kg, NTP could ask:

• Does incidence rise with exposure?
• Is the relationship monotonic?
• Does an intermediate dose produce the largest response?
• Do neoplastic and preneoplastic lesions move together?
• Does the same pattern appear under more than one modulation?
• Is a response concentrated in one sex or tissue?

With only 0 and 4 W/kg, the Japan–Korea design could ask only:

Did this particular sample produce a statistically significant pairwise difference under this one CDMA condition?

That is a legitimate but narrow question.

It cannot answer whether RF is carcinogenic across doses, whether a response exists below 4 W/kg, whether 6 W/kg reproduces the NTP heart result, or whether GSM produces a different pattern.

Melnick and colleagues summarize the methodological principle clearly: a follow-up experiment intended to refute a positive carcinogenicity study must duplicate the relevant conditions or exceed the original study’s sensitivity. It must include the doses at which the original response was significant and the doses needed for trend analysis.

The Japan–Korea studies did neither.

---

The ICNIRP connection is relevant—and should be stated accurately

Young Hwan Ahn was a corresponding author on the Korean paper and a lead author of the project’s 2022 design publication. ICNIRP’s official biography says that he participated in the Japan–Korea NTP-validation study and joined ICNIRP’s Main Commission in July 2024, while the project was underway.

The new Melnick–Moskowitz–Héroux commentary also reports that the project’s principal investigator is an ICNIRP member and that the project’s international advisory committee included substantial representation from current and former ICNIRP members.

This does not, by itself, establish fraud, data manipulation, or personal wrongdoing.

It does establish a consequential alignment of institutional framework:

• ICNIRP’s thermal threshold supplied the single selected dose.
• ICNIRP-affiliated experts participated in the project or its oversight.
• The study’s broad interpretation was then used to reinforce the adequacy of ICNIRP’s thermal-based limits.

That is exactly the kind of setting in which complete disclosure, independent replication, prospective power calculations, and strict limits on interpretation are essential.

No accusation of conspiracy is needed. The methodological circularity is visible in the published record.

---

The low-dose evidence was not tested

The Ramazzini Institute exposed 2,448 Sprague–Dawley rats from prenatal life until natural death to 1.8 GHz GSM base-station-like fields for 19 hours per day. Estimated whole-body SARs ranged from approximately 0.001 to 0.1 W/kg—far below NTP’s levels and far below the 4 W/kg Japan–Korea condition. Ramazzini reported a statistically significant increase in heart schwannomas in exposed males at its highest exposure, along with increases in Schwann-cell hyperplasia and malignant glial tumors that did not individually reach significance.

The Japan–Korea studies did not test that exposure range.

They therefore cannot resolve the central observation that similar rare tumor types appeared in two independent lifetime studies using:

• different exposure facilities;
• different frequencies;
• near-field and far-field conditions;
• and SARs separated by orders of magnitude.

A 2025 WHO-commissioned systematic review concluded that the certainty of animal evidence was strongest for malignant heart schwannomas and gliomas. That assessment has been challenged by ICNIRP-aligned commentators, but the existence of that debate makes a properly powered, multi-dose replication more—not less—important.

The underpowered 4 W/kg studies did not settle that debate.

---

Why the “no cancer in rats” headline is scientifically indefensible

ARPANSA’s headline says the Korean and Japanese studies showed that RF “does not cause cancer in rats.”

But the evidence actually supports a much narrower sentence:

In two male-rat studies using one 900 MHz CDMA exposure level of 4 W/kg, no tested tumor comparison reached conventional statistical significance.

That sentence is accurate.

It is also dramatically less reassuring than the headline because it preserves the missing information:

• only males were studied;
• only CDMA was studied;
• only one exposed SAR was studied;
• no dose trend was possible;
• the groups were underpowered for the rare target tumors;
• the NTP 6 W/kg CDMA condition was omitted;
• the NTP GSM condition was omitted;
• lower Ramazzini-range exposures were omitted;
• and Korea still found two heart schwannomas only in exposed rats.

The difference between those two sentences is the difference between reporting a study and advertising an institutional conclusion.

---

Did the Japan–Korea work strengthen NTP?

It did not statistically replicate NTP, and it should not be described as proof of the NTP result.

But it also did not produce the opposite biological pattern.

At the most important Korean endpoint, it produced the same rare tumor exclusively in the exposed group:

$$0/70,0/70,2/70.$$

That 2.9% incidence closely matches the NTP CDMA incidence at 3 W/kg. Given the low statistical power, the result is compatible with both a modest exposure-related effect and random occurrence. What it is not compatible with is the claim that the Korean experiment documented a complete absence of the NTP-associated tumor.

The strongest defensible interpretation is:

The Korean result is directionally supportive, statistically inconclusive, and incapable of refuting NTP.

The Japanese result—one heart schwannoma in each of the cage, sham, and RF groups—does not show the Korean pattern. But with a single dose and low power, neither national study can reliably adjudicate a rare-tumor response expected to involve only a few animals.

Taken together, the studies strengthen the case for a genuinely decisive experiment. They do not strengthen the claim that NTP was wrong.

---

What a real validation study would require

A scientifically decisive follow-up should be designed around the original findings rather than around an existing guideline threshold.

It should include:

1. Both GSM and CDMA, preserving the complete exposure waveforms and making them publicly available.
2. Multiple SAR levels, including at minimum NTP’s 1.5, 3, and 6 W/kg conditions, plus lower far-field conditions spanning the Ramazzini range.
3. Prospective rare-tumor power calculations, with enough animals to achieve at least 80%—preferably 90%—power for heart schwannomas and gliomas.
4. Both sexes, because sex-specific sensitivity was part of the NTP result and cannot be investigated in male-only studies.
5. Concurrent cage and sham controls, blinded pathology, mortality-adjusted statistics, pairwise tests, and trend analysis specified before unblinding.
6. Time-resolved mechanistic measurements, including body temperature, calcium dynamics, mitochondrial function, oxidative stress, antioxidant responses, heat-shock signaling, oxidative DNA lesions, apoptosis, senescence, and immune markers.
7. Factorial waveform testing, separating carrier frequency, modulation, duty cycle, peak-to-average structure, cumulative exposure, and absorbed energy.

Only such a design could determine whether the response is:

• monotonic;
• nonmonotonic;
• modulation-specific;
• sex-specific;
• thermally mediated;
• non-thermally mediated;
• or shaped by competing injury and adaptation processes.

Until then, the boundary-layer question remains open.

---

This matters because exposure limits were never based on cancer risk

In a separate March 2026 analysis, Melnick and Moskowitz applied benchmark-dose and conventional public-health risk-assessment methods to experimental cancer and reproductive data. Their model estimated that current public whole-body limits are between 15 and 900 times higher than exposure levels associated with an added cancer risk of one in 100,000, depending on daily exposure duration, and 8 to 24 times higher than levels they estimated would protect male reproductive health. Those are model-based estimates rather than direct measurements of human risk, but they demonstrate why the animal evidence cannot be dismissed without an adequate replacement study.

Current limits were developed to prevent acute thermal disruption—not to control chronic cancer risk, developmental effects, fertility loss, or waveform-dependent biological perturbation. Choosing the thermal threshold as the sole dose in a cancer-validation study and then using the resulting null to defend thermal limits is not an independent test of those assumptions.

It is a loop.

And it is precisely the loop that independent science must break.

---

The correct conclusion

The Japan–Korea studies can be accurately described as:

Well-executed, harmonized, GLP-compliant, single-dose studies of 900 MHz CDMA exposure at 4 W/kg in male rats.

They cannot accurately be described as:

• complete replications of NTP;
• tests of NTP’s GSM findings;
• tests of NTP’s full CDMA dose response;
• tests of lower-dose Ramazzini findings;
• demonstrations that RF cannot cause cancer;
• or validations of the safety of ICNIRP and FCC limits.

The new commentary by Melnick, Moskowitz, and Héroux makes the methodological failure unmistakable: the follow-up studies were smaller, used only one exposed dose, could not perform trend analysis, omitted the NTP dose with the strongest CDMA heart result, and were severely underpowered for the rare tumors they were supposed to evaluate.

The Korean study even found the signature heart tumor in two exposed animals and none in either concurrent control group.

That is not proof.

But it is certainly not a clean bill of health.

The Japan–Korea studies did not rebut NTP. They demonstrated what happens when a broad biological question is forced through a narrow, guideline-centered experimental window: the result becomes a boundary-layer null that cannot resolve the science—but can still be misused to create a reassuring headline.

The honest scientific verdict is therefore:

NTP remains unrebutted.

And until researchers conduct a properly powered, multi-dose, multi-modulation experiment that actually tests the conditions where the original signals appeared, no responsible assessment should claim otherwise.