Overcorrection

Mechanism #26

Overcorrection Oscillation

In 2002, the Women's Health Initiative trial was stopped early. Hormone replacement therapy, prescribed to millions of women, appeared to cause breast cancer and heart disease. The FDA added its most severe warning. Prescriptions dropped by up to 80%. Doctors refused to write them. Women stopped asking.

The trial had enrolled women averaging 63 years old — a decade or more past menopause. Most women start HRT at 45 to 55. Only 30% of participants were in the age range where the treatment is actually used. The study tested the wrong population and generalized the results to everyone.

Philip Sarrel and colleagues at Yale estimated the cost: between 18,601 and 91,610 premature deaths among US women aged 50 to 59 between 2002 and 2012, from avoiding estrogen therapy that would have reduced their all-cause mortality by 39%. Robert Langer, lead investigator at a WHI vanguard site, later called the trial's early termination and reporting "highly unusual circumstances" that resulted in "misinformation and hysteria."

In November 2025, twenty-three years later, the FDA removed the warning. HHS called it "restoring gold-standard science." For women under 60 starting within ten years of menopause, reanalysis showed HRT reduces cardiovascular disease by up to 50%, Alzheimer's by 35%, fractures by 50 to 60%. A 2026 Danish study of 800,000+ women found menopausal HRT not associated with increased death risk.

Then former WHI safety leaders warned that the removal "could put millions of women at risk." They called for "a true, transparent advisory panel before this rushed overhaul."

Both sides have dead women on their ledger. Neither side counts the other's.

This is not a story about hormones. This is the pattern.

The Architecture of Overshoot

When a system discovers genuine error, the correction overshoots the optimum. Not by accident — by architecture. The visible harm from the original error weighs roughly twice the invisible harm from the correction (Kahneman and Tversky's loss aversion, applied to policy). The correction then entrenches through the same institutional inertia that protected the original error: new guidelines, new training defaults, new medico-legal norms. And the victims of the overcorrection are classified as "people who didn't receive treatment" rather than "victims of the correction" — so no one counts them.

The system oscillates between positions rather than converging on truth. Each swing creates its own invisible graveyard.

Prasad and Cifu, analyzing a decade of NEJM publications, found that 40.2% of tested established medical practices were subsequently reversed. They mapped the reversal itself. What happens after — the dynamics of the correction — is unmapped territory. That's this mechanism.

Opioids: The Six-Year Correction Cycle

The original crisis was real. Liberal prescribing, fueled by Purdue Pharma's marketing of OxyContin and a pain-management movement that treated pain as a "fifth vital sign," produced an epidemic. Hundreds of thousands dead.

In 2016, the CDC issued its Guideline for Prescribing Opioids for Chronic Pain. Intended as flexible recommendations, they were implemented as rigid law. Insurers imposed hard dosage ceilings. Pharmacists refused prescriptions. Physicians force-tapered patients who had been stable on opioids for years — sometimes overnight. The guidelines were extended to populations they were never designed to cover: cancer patients, palliative care, post-surgical recovery.

The counter-harm: untreated pain, forced withdrawal, psychological distress. Multiple observational studies linked abrupt opioid discontinuation to increased rates of overdose, emergency department visits, and suicide. Nine percent of suicide decedents had evidence of chronic pain at time of death. Black patients were disproportionately discontinued — the overcorrection amplified existing inequities.

In 2019, the FDA acknowledged increased suicide risk from abrupt opioid discontinuation and mandated label changes. In 2022, the CDC updated its guidelines to explicitly state: "Do not set rigid dosage thresholds."

Meanwhile, 105,007 Americans died from opioid overdoses in 2023 alone — mostly from illicit fentanyl. The original crisis didn't go away. The correction shifted who was harmed: from people given too many prescription opioids to people denied adequate pain treatment, while the street supply continued unabated.

Dietary Fat: The Fifty-Year Oscillation

In the 1960s, the sugar industry paid Harvard scientists to shift blame from sugar to saturated fat. In 1968, the AHA recommended no more than 300mg of cholesterol per day and no more than three eggs per week. Low-fat dietary guidelines dominated public health for the next half-century. The food industry responded by producing low-fat products that replaced fat with sugar and refined carbohydrates.

The result: minimal reduction in fat consumption, massive increase in calories and sugar. The obesity rate tripled over forty years. Heart disease didn't decline as projected.

In 2015, the Dietary Guidelines Advisory Committee dropped the 300mg/day cholesterol limit. "Cholesterol is not a nutrient of concern for overconsumption." The same document also said: "Eat as little dietary cholesterol as possible." The contradiction within the correction.

The 2025–2030 guidelines finally acknowledge that ultra-processed foods, not fat itself, are the primary concern. But the saturated fat cap persists at 10% — and the counter-narrative has its own problems. Columbia and CUNY historians have challenged the "sugar conspiracy" framing, arguing the Harvard scientists were already pro-low-fat before industry funding arrived. Even the story about the overcorrection may itself be overcorrected.

Tonsillectomy: The Surgical Pendulum

In the 1950s and 1960s, tonsillectomy was massively overperformed. Some physicians said the indication was simply "the presence of tonsils." By the 1970s, backlash drove rates down by 75%.

By 2018, the pendulum had swung past center. UK data showed that of 15,760 children with sufficient recurrent sore throats to clinically justify tonsillectomy, only 13.6% — 2,144 children — received surgery. Meanwhile, pediatric obstructive sleep apnea (5.7% incidence) went undertreated, carrying risks of behavioral problems, poor academic performance, and cardiovascular complications.

Forty years from overcorrection to recognized under-treatment. The victims: children classified as "not needing surgery" rather than "denied surgery by institutional backlash."

Deinstitutionalization: The Morally Correct Overcorrection

This case is different. The correction was right.

In 1955, 558,239 people occupied state psychiatric hospitals — 340 beds per 100,000 population. Fifty percent of all hospital beds in the United States were psychiatric beds. The institutions were often genuinely abusive: overcrowded, understaffed, punitive.

Kennedy signed the Community Mental Health Centers Act in 1963. Only 700 of the planned 1,500 community centers were ever built. But the hospitals closed anyway — incentivized by Medicaid amendments that shifted costs from states to the federal government. By the 2020s, 95% of psychiatric beds were gone. 558,239 patients became approximately 45,000. Bed capacity fell to ~14 per 100,000 — back to 1850 levels.

The Penrose effect, documented across multiple studies: for every psychiatric bed removed per 100,000 population, 0.8 people with serious mental illness were incarcerated. Correlation between bed decline and incarceration: R = −0.53. Between bed decline and homelessness: R = −0.71.

Today, roughly 378,000 people with serious mental illness are incarcerated. Twenty to thirty percent of the homeless population has serious mental illness. The correction was morally right in direction — institutions were abusive. The overshoot was in speed and completeness: the system closed hospitals before community alternatives existed. The patients didn't go home. They went to the streets and to jail.

The Counter-Case: Vioxx and Partial Convergence

Not every correction oscillates. When Merck's Vioxx (rofecoxib) was withdrawn in 2004 — after causing an estimated 88,000 to 139,000 heart attacks — the FDA did something unusual: it differentiated within the drug class.

Celecoxib, a related COX-2 inhibitor, was allowed to remain on the market with additional warnings. The distinction: celecoxib was the least selective COX-2 inhibitor. The cardiovascular harm mapped to a distinguishable molecular property. The FDA mandated the PRECISION trial (2016, n=24,081), which found celecoxib no worse than ibuprofen or naproxen for cardiovascular events. Vindication — twelve years later.

But the correction was only partially proportional. Class-wide suspicion drove many patients to traditional NSAIDs (which carry their own cardiovascular risk, never rigorously tested before Vioxx). Scottish prescribing data showed celecoxib initially rose, then fell as class-wide fear spread. The overcorrection was real — just smaller and shorter-lived than HRT or opioids.

What was different? The harm mapped to a property that could be measured at the molecular level. The system could tell one drug from another. It could target the correction.

What Separates Oscillation from Convergence

Five factors predict whether a correction will oscillate or converge. The pattern holds across all six cases:

Factor	Oscillates	Converges
Target	Entire class or system	Specific instance
Distinguishability	Members of class treated identically	Measurable differences within class
Moral framing	Safety / rights / epidemic	Evidence-contingent
Feedback latency	Years to decades	Rapid
Counter-harm visibility	Victims classified away	Victims identifiable

The fifth factor — counter-harm visibility — connects this mechanism directly to counterfactual invisibility (mechanism #25). The less visible the victims of overcorrection, the longer the oscillation persists. HRT victims were invisible for eleven years because they were counted as "women who didn't take hormones," not as "women harmed by a policy." The classification hides the causation. The mechanisms interact.

The Oscillation Continues

Bastiat saw this in 1850: the seen versus the unseen. Loss aversion makes it structural. Moral framing makes it durable. Institutional inertia makes it self-reinforcing.

The FDA just removed a 23-year-old warning. HHS called it "restoring gold-standard science." Former WHI investigators called it "rushed." Menopause influencers are calling HRT a miracle. Oncologists are calling for caution. The evidence sits in the same place it's sat for a decade — but the oscillation generates its own momentum, independent of the data.

The pattern predicts what happens next: the pendulum swings back, new guidelines overcorrect in the other direction, and a new generation of invisible victims is classified into existence.

Vinay Prasad, who mapped medical reversals, found that 40% of established practices are eventually overturned. He documented the what. This mechanism maps the after — the structural reason that corrections don't converge on truth but oscillate through it, leaving graveyards on both sides of the optimum that no one audits because the dead are always classified as something other than what killed them.