480. My doubts about Parkinson’s medications: Levodopa and others, part one.

READER RESOURCES: All my books are now available as paperbacks on Lulu. THE APOCALYPSE ALMANAC: Hidden cures in our dystopian age. Check out the “Cure Cancer in Your Kitchen” chapter. FULLSCRIPT SUPPLEMENTS: top quality and economical. AFFILIATE STORE: I offer competitively priced products HERE that I have personally tested and used.

Preface

I know what you’re thinking. What the heck is Yoho doing with this overly long exposition on Parkinson’s drugs? The answer is that I am explaining my suspicions, based on my background. This is critical information for those who have Parkinson’s or family members with Parkinson’s.

Neurologists have very few therapies that work well. Levodopa is a drug described as having a “honeymoon period.” This means it invariably reduces the body’s responsiveness over time, which must be damaging at some level.

Their special Parkinson’s scoring system, part 4, says “Dyskinesia refers to involuntary, erratic writhing movements caused by long-term medication use” (my emphasis). This refers to levodopa. But admitting that levodopa is flawed is too much of a stretch for them.

Last month, a neurologist told me that levodopa and related drugs reduce mortality in Parkinson’s disease. I had never heard this in 40 years of medicine. Every other neurologist I have met has recited the same catechism: the drugs treat symptoms and do not alter the disease course. That statement is a hedge dressed as reassurance. A drug that alters nothing would be rare in pharmacology, and a profession that repeats the phrase for half a century is either hiding a benefit it never proved or a harm it never looked for.

I have spent 10 years documenting how Pharma manufactures evidence, and I laid out its methods in Butchered by “Healthcare.” These include relative risk without absolute risk, surrogate endpoints in place of death, after-the-fact subset analyses, and trials with casts of thousands designed to detect differences too small to matter. The Parkinson’s literature uses every one of these devices, and the confounds are numerous and profound. There is even occasional fraud. What follows is my reading of the mortality evidence, with the studies named so a reader can check them.

I say what I have concluded about these drugs at the end of the essay, but if you, a friend, or a family member has Parkinson’s, you should study this and try to figure it out yourself. All the information you need to make your decision is here in this (unfortunately) 7,000-word essay. No one is coming to save any of us, and there is a lot of medical bullshit here to sort through.

I use some of the technical language from the trial literature. None of it is difficult, and the handful of terms that matter are defined in a short section following the summary below.

Summary

Before levodopa, Parkinson’s patients died at 2.9 times the rate of people their age. The first treated cohorts, followed from 1968 to 1975, reported ratios of 1.3 to 1.9. That comparison, treated patients of one decade against untreated patients of an earlier one, is the sole basis for the claim that the drug prolongs life. Every cohort followed for more than 10 years shows the benefit fading. Innsbruck’s standardized mortality ratio went from 0.6 at 5 years to 2.0 at 38 years; Sydney’s reached 3.1 at 20 years; and Ontario found a 2.5-fold death rate among 15,304 treated patients.

No randomized trial in 56 years has tested whether levodopa reduces mortality. The largest trial randomizing initial therapy, PD MED, found no survival difference among levodopa, agonists, and MAO-B inhibitors after 15 years. The only randomized mortality signal involving levodopa was harmful: British patients given levodopa plus selegiline died 57 percent faster than those given levodopa alone. A 2024 Dutch analysis of 29,943 patients found that early levodopa neither shortened nor lengthened life, and a 2014 review of 88 studies found no evidence that mortality declined after the drug’s introduction. Meanwhile, American Parkinson’s death rates rose by 63 percent between 1999 and 2019, during the era of universal dopaminergic therapy.

The evidence that the drug destroys neurons proved weaker than I expected. Brain-bank studies find no link between lifetime levodopa dose and nigral cell loss, and the loss of drug response over time tracks disease duration in untreated Africans on the same schedule as in treated Westerners. The drug’s failure is different: it does nothing to the disease, and a drug that does nothing to a fatal disease cannot change how it ends.

The one intervention with a mortality figure worth reporting is not a drug. Moderate exercise before and after diagnosis cut mortality in half in a Korean cohort of 10,699 patients. No medication has matched that effect.

A short guide to the technical terms

A few terms recur throughout what follows, and a reader who has them in hand will find the rest of the essay easier.

Relative risk is the death rate in one group divided by the death rate in another. A relative risk of 2 means twice as many deaths. A hazard ratio is the same idea computed over time, and an odds ratio is a close cousin used in some study designs; for a rare-enough outcome, all 3 read the same way. Absolute risk is the plain percentage of people who died, and the difference between absolute and relative risk is the subject of a chapter in my book: a drug that cuts deaths from 2 percent to 1 percent has halved the relative risk while sparing 1 patient in 100, statistical fraud used over and over again in academic medicine.

A standardized mortality ratio compares the number of deaths observed in a group of patients with the number expected in people of the same age and sex drawn from the general population. A ratio of 1.0 means the patients died at the ordinary rate, 2.0 means twice the ordinary rate, and 0.6 means they died less often than their neighbors. Excess mortality is the same comparison expressed as a surplus.

A confidence interval is the range within which the true value plausibly lies, given the data. When a confidence interval for a ratio spans 1.0, the study has not shown a difference in either direction. A p-value is the probability of observing a result at least this large if the drug had no effect; by convention, a p-value under 0.05 is considered significant, which says nothing about whether the result matters to a patient.

A randomized trial assigns patients to treatments by chance, so the groups start alike, and it is the only design that separates what a drug does from who received it. An observational or cohort study follows patients who receive treatment in the usual way, and its findings are open to 3 errors that recur in this essay. Confounding by indication means the drug went to a particular kind of patient, so the outcome reflects the patient rather than the pill. The healthy-user effect means people who take medicine faithfully live longer, regardless of the medicine. Immortal time bias means a patient must survive long enough to be counted as a user, so users look longer-lived by definition.

The UPDRS, the Unified Parkinson’s Disease Rating Scale, is the standard clinical score; higher scores indicate worse disease, and a few points represent a small change. SPECT imaging with a tracer such as beta-CIT measures the dopamine transporter in the striatum, a proxy for surviving dopamine nerve terminals. A delayed-start trial gives one group the drug immediately and the other after a delay, and asks whether the late starters ever catch up; if they do not, the drug is said to have modified the disease. A washout is the period after a drug is stopped before measuring its effect.

The number 2.9

Margaret Hoehn and Melvin Yahr published their natural history study in Neurology in 1967, the year before levodopa reached patients. They followed 802 patients seen at Columbia between 1949 and 1964. The observed death rate was 2.9 times the expected rate for people of the same age and sex. About a third of untreated patients were dead or severely disabled within 5 years, and more than 80 percent were by 15 years. Their five-stage scale is still used in every clinic.

That single ratio, 2.9, became the yardstick for every subsequent survival claim. Any treated group that died at less than 2.9 times the population rate was claimed to have benefited from the drug. Hold onto that comparison, because it is a before-and-after comparison across decades, hospitals, diagnostic standards, and patients, and it underpins the whole shaky edifice.

The first survival claims

George Cotzias, a Greek-born physician at Brookhaven, gave patients oral D, L-dopa in doses up to 16 grams per day and reported the results in the New England Journal of Medicine in 1967. Some patients developed dangerous drops in white blood cell counts. He switched to the pure L-form, titrated it upward over weeks, and reported on 28 patients in 1969. The response was unlike anything neurology had seen. Rigid, frozen patients stood up and walked. He won the Lasker Award that year, and the FDA approved levodopa in 1970. Merck’s carbidopa combination, Sinemet, followed in 1975. Cotzias died of lung cancer in 1977, 3 days short of his 59th birthday.

In 1969, Oliver Sacks gave the same drug to post-encephalitic patients at Beth Abraham in the Bronx and wrote Awakenings about it in 1973. His patients came back to life, then developed wild swings, tics, tolerance, and collapse within months. His letters to the journals describing the failures were met with skepticism from colleagues who wanted only the miracle. A December 2025 New Yorker article by Rachel Aviv, drawn from his journals, reported that Sacks gave his patients abilities they did not possess and called some of his own details “pure fabrications.” He wrote that he had misstepped many times and that the book was a source of severe, long-lasting self-recrimination. Anyone who has read the book will recognize that its second half, the part about the drug’s failure, was the truer half.

The first survival reports appeared within a few years. Sweet and McDowell followed 100 patients who started levodopa before the end of 1968 and, in the Annals of Internal Medicine in 1975, reported that 32 had died at 5 years, an excess mortality of 1.9 compared with the American population. Only 47 of the 100 were still taking the drug. Functional ratings peaked at 1 to 2 years and then declined toward pre-treatment levels. Joseph and colleagues reported on a Roche-sponsored multicenter cohort in 1978, with a death rate about 33 percent above the general population, a ratio near 1.3. Diamond and Markham combined data from 359 patients across 4 centers in 1987 and found that those treated within 3 years of onset had an observed-to-expected ratio of 1.43, compared with 2.44 for those treated at 4 to 6 years and 2.95 for those treated at 7 to 9 years.

These 3 papers, all comparing treated patients from the 1970s with untreated Columbia patients from the 1950s, make the claim that levodopa prolongs life. Every later claim descends from them.

The comparison has a problem that the authors themselves acknowledged. Patients enrolled in the 1968 to 1970 drug trials were fit enough to travel to a research center, tolerate weeks of titration, and remain in follow-up. Hoehn and Yahr’s patients included the bedridden and the institutionalized from an earlier era. Scigliano and colleagues tested the early-versus-late idea in 145 Italian patients in 1990 and found that late starters died at 2.5 times the rate of early starters. After adjusting for age, the difference disappeared. Disease severity at baseline predicted death by a factor of 4, and the timing of the drug predicted nothing.

The long cohorts and the vanishing benefit

The Sydney Multicentre Study enrolled 149 new patients in the 1980s and followed them for 20 years. At 10 years, the standardized mortality ratio was 1.58. At 15 years, it was 1.86, and 95 percent of survivors had dyskinesia. At 20 years, 100 of the 136 confirmed cases had died, and the ratio in the last 5-year interval was 3.1, higher than Hoehn and Yahr’s untreated figure. Of the survivors, 83 percent were demented, 87 percent fell, 81 percent froze, and 48 percent lived in nursing homes. Hely’s group titled the 15-year paper to state that “non-L-dopa-responsive problems dominate at 15 years.”

The Innsbruck cohort provides the clearest picture. Diem-Zangerl and Poewe followed 238 patients with onset between 1974 and 1984. The standardized mortality ratio was 0.6 at 5 years, meaning treated patients died at little more than half the expected rate. It was 0.9 at 10 years, 1.2 at 15, and 1.3 at 20. Pinter extended the cohort to 38 years in 2015, by which time 230 of 237 were dead and the ratio was 2.02. Pneumonia, stroke, and heart disease accounted for the excess.

That early ratio of 0.6 is the strongest evidence in the literature that the drug affects survival. It is also the honeymoon. During the first years after diagnosis, a levodopa-responsive patient with verified disease walks, swallows, and stays out of bed, and he dies less often than his neighbors because clinic patients are selected for being well enough to attend. As the disease reaches parts of the brain the drug does not reach, the ratio climbs, and by 20 to 40 years the treated cohort has caught up with the untreated one.

Population studies say the same thing, without the flattering early years. Guttman’s Ontario study matched 15,304 Parkinsonian patients with 30,608 controls over 6 years. Half the patients died, compared with 29 percent of controls, yielding an odds ratio of 2.5. The authors wrote “despite modern drug therapy” in the title. Driver’s analysis of 560 physicians in the Physicians’ Health Study found a hazard ratio of 2.32, present within 2 years of diagnosis. Fall’s Swedish cohort found a hazard ratio of 2.4. Elbaz’s Olmsted County incident cases found a hazard ratio of 1.6. Xu’s 2014 meta-analysis of 8 prospective studies with 72,833 participants found a pooled relative risk of 2.22, rising to 3.78 with dementia. A Korean national study published in 2023 compared 8,220 patients with 41,100 controls in a fully treated era and found a hazard ratio of 2.96, indistinguishable from that in 1967.

Macleod, Taylor, and Counsell reviewed 88 studies in Movement Disorders in 2014. Mortality ratios ranged from 0.9 to 3.8. In inception cohorts, the pooled ratio was near 1.5. Mortality rose over time in every study, and they wrote that no evidence indicated decreased mortality after the introduction of levodopa. That is the largest systematic review on the question, and it is the sentence the neurologists leave out.

The trials that were never run

Death is the hardest of endpoints and can never be disputed by anyone who can count bodies. My book’s chapter on easy math says it in one line: if the death rate is unchanged, the patient benefit is doubtful. Levodopa has been on the market for 56 years, and no randomized trial has compared it with no treatment for mortality. The reason given is that withholding it would be unethical. That is the same argument used for every drug the industry wants left untested, and it is circular, since the ethical case rests on a survival benefit that no one has measured.

What the trials measured is instructive.

ELLDOPA, published by Fahn and the Parkinson Study Group in the New England Journal in 2004, randomized 361 early patients to placebo or to 150, 300, or 600 milligrams of levodopa daily for 40 weeks, followed by a 2-week washout. Clinically, the drug won. The placebo group worsened by 7.8 points on the UPDRS scale, and the 600-milligram group improved by 1.4 points. Then the imaging substudy of 142 patients showed the opposite. Dopamine transporter uptake on SPECT declined 1.4 percent on placebo and 7.2 percent on 600 milligrams. The investigators wrote that the clinical data suggested the drug slowed the disease, while the imaging suggested it accelerated the loss of nerve terminals, and they did not know which was true. The trial was designed to settle the toxicity question, and it settled nothing. The 2-week washout was too short to remove the drug’s long-duration effect, so the clinical result is contaminated; SPECT tracer binding is also affected by dopamine levels, so the imaging result is contaminated too. Twenty-two years later, it has never been repeated with a longer washout.

LEAP, the Dutch delayed-start trial published in 2019, randomized 445 patients to receive 300 milligrams daily, either immediately or after 40 weeks. At 80 weeks, the difference was 1.0 point on the UPDRS, with a confidence interval that spanned zero. The 5-year follow-up, published in 2024, found no difference either, and dyskinesia was numerically higher in the early group. The authors concluded that levodopa has no disease-modifying effect. No one has explained how a drug that neither helps nor harms the disease process is supposed to change how long a patient lives once the honeymoon ends.

PD MED was the trial the British public paid for so industry would not. In the 2014 Lancet report, Gray and Clarke randomized 1,620 patients to start with levodopa, a dopamine agonist, or an MAO-B inhibitor and followed them for up to 7 years. Levodopa improved quality of life by 1.8 points on a 100-point mobility scale. Mortality did not differ: the hazard ratio was 0.85, with a confidence interval from 0.69 to 1.06. The final 15-year analysis, presented in 2023, found no evidence of any difference in mortality among the 3 strategies. If levodopa extended life through some direct action on survival, a trial that gave 528 patients levodopa from day one and 1,092 patients something weaker should have found it. It found nothing.

The one randomized mortality signal is the one nobody advertises. The UK Parkinson’s Disease Research Group randomized 520 untreated patients across 93 hospitals to receive either levodopa alone or levodopa plus selegiline. Lees reported in the BMJ in 1995 that the combination group died faster, with a hazard ratio of 1.57 and a confidence interval from 1.09 to 2.30. The 1998 follow-up, with 624 patients and 120 deaths, trimmed the excess to 1.32, with Parkinson’s-specific deaths at 2.5 times the rate. The manufacturer’s response was a 1998 Neurology meta-analysis by Olanow, pooling 5 trials and finding no excess, and a general-practice database study by Thorogood, finding a hazard ratio of 1.11. The prescribing literature treats the matter as refuted. A 57 percent excess death rate in a government-funded randomized trial of 520 patients was answered with observational data and a pooled analysis of trials never designed to count deaths. My book has a chapter on how this sort of fraud is perpetrated.

The other trials did not report mortality because they were not designed to. DATATOP enrolled 800 untreated patients and found that neither selegiline nor vitamin E changed the death rate over 8 years, which was 2.1 percent per year. CALM-PD compared pramipexole with levodopa and reported dyskinesia rates and SPECT decline, but nothing on survival. STRIDE-PD added entacapone to levodopa in 747 patients and produced more dyskinesia, sooner, and a cardiovascular and prostate cancer imbalance that the FDA reviewed in 2010. ADAGIO, the rasagiline delayed-start trial in 1,176 patients, met its endpoint at 1 milligram and failed it at 2 milligrams, a result no pharmacologist has explained, and reported no deaths as an outcome.

Where the new claim comes from

I looked for the paper my acquaintance had in mind. No 2024, 2025, or 2026 publication shows that levodopa, compared with no levodopa, lowers mortality. The closest candidates are these.

Talebi and colleagues used Dutch insurance claims covering 29,943 new Parkinson’s patients and applied marginal structural models, the current statistical approach for extracting causal effects from observational data. Early levodopa initiators numbered 24,847, and non-early initiators numbered 5,096. The causal risk ratio for death was 1.04 (95% CI, 0.92 to 1.19). Early levodopa neither postponed nor accelerated death. This is the largest modern analysis of the question, and it found no effect.

Lee and colleagues used the Korean national insurance database to follow 4,949 new patients for 5 years, and published the study in 2026. Compared with levodopa alone, adding an MAO-B inhibitor early was associated with a mean survival gain of 3.45 months, adding an agonist with 0.95 months, and adding a COMT inhibitor with 1.64 months less. The authors wrote that the results likely reflected disease severity and prescriber preference. This is the kind of paper that becomes “dopaminergic therapy reduces mortality” by the time it reaches a dinner party.

A Swedish registry study of 21,047 patients found that those started on agonists survived longer than those started on MAO-B inhibitors. Because agonists are prescribed to younger, fitter patients, the study shows who received the drug, not what the drug did.

Barer and colleagues, in an AbbVie-funded Israeli study of 2,525 patients, found that a levodopa-equivalent dose above 600 milligrams in the first year predicted a 50 percent higher death rate, and an 80 percent higher rate in patients under 75. They interpreted the dose as a marker of severity rather than toxicity. That is the correct interpretation, and it cuts both ways: if a high dose indicates a sick patient rather than a harmful drug, then a low dose indicates a well patient rather than a helpful drug, and the entire early-treatment literature dissolves.

A 2025 Neurology study screened 219 drugs among 14,289 Parkinson’s patients and identified 23 linked to lower 8-year mortality, including tadalafil, losartan, and atorvastatin. It excluded antiparkinsonian drugs by design. I suspect this study, garbled in transmission, is the source of the rumor.

Why the observational studies flatter the drug

Three biases produce a survival benefit out of nothing, and all 3 operate within this literature.

Confounding by indication is the first. Neurologists give levodopa to patients who respond to it, and response to levodopa is a feature of the disease rather than the drug. Marras analyzed the DATATOP cohort over 13 years and found that a poor levodopa response predicted death. A patient with atypical parkinsonism, which kills in 6 to 8 years and responds to nothing, is counted as a non-responder or a non-user. The drug gets credit for the diagnosis.

The healthy-user effect is the second. People who take their pills, keep their appointments, and stay in studies live longer, regardless of what the pills contain. This was demonstrated in the Coronary Drug Project half a century ago, when men who faithfully took their placebo died at half the rate of those who did not.

Immortal time bias is the third. A registry defines a levodopa user as someone who filled 3 prescriptions over 6 months. A patient who dies in month 4 is counted as a non-user. Viuff pointed out in 2026 that the Norwegian registry definitions build exactly this error into drug-repurposing analyses. This bias creates a survival advantage for any exposure that takes time to accumulate.

Against this machinery, Hoehn and Yahr’s 2.9 offers no defense. It is a number from a different century, and the modern cohorts, once the honeymoon ends, land on top of it.

The dopamine agonists

The agonists were marketed as levodopa-sparing, a term implying levodopa toxicity even as the same companies denied it. Their record is the worst in the class.

Pergolide and cabergoline are ergot derivatives that stimulate serotonin receptors on heart valves. Schade and Suissa reported in the New England Journal in 2007 that pergolide increased the rate of valve regurgitation 7-fold and cabergoline 5-fold among 11,417 British patients. Zanettini’s echocardiography study in the same issue found moderate-to-severe regurgitation in 23 percent of pergolide patients and 29 percent of cabergoline patients, compared with 6 percent of controls. Permax was withdrawn from the American market in March 2007, 18 years after approval, despite heart damage in thousands of patients that a single echocardiogram in the approval trials would have detected.

Pramipexole and ropinirole, non-ergot agonists, replaced them. Weintraub’s DOMINION study screened 3,090 patients at 46 centers in 2010 and found impulse-control disorders in 17 percent of agonist users, compared with 7 percent of non-users. Pathological gambling, compulsive shopping, hypersexuality, and binge eating destroyed marriages and bank accounts. Boehringer Ingelheim knew of gambling reports for years before adding them to the Mirapex label in 2005. The first American jury awarded $8.3 million in 2010 to a man who lost $260,000 gambling. Frucht reported in 1999 that 8 patients on pramipexole and 1 on ropinirole fell asleep at the wheel and crashed. The companies called these sleep attacks, a new term for an old side effect.

No trial has shown that an agonist reduces mortality. PD MED found that patients started on agonists had worse quality of life than those started on levodopa and died at the same rate.

Parkinson’s is so common that you probably know a lot of people who should read this post. As I understand it, Substack now allows free subscribers to comment on paid posts like this.