I have spent nearly twenty-five years treating osteoporosis. The drugs have gone from oral and intravenous bisphosphonates to several classes with better efficacy and, mostly, better tolerability. Their reputation has gotten worse. Some of that is a jaundiced view of the pharmaceutical industry, some the internet doing what the internet does. But a meaningful share comes from the government. The Food and Drug Administration (FDA) prints warnings its own trial data do not support. Blogs, forums, and influencers deliver them to my exam room as a reason to decline treatment.
Osteoporosis is silent until it is not. Convincing a woman with a T-score of −3.0 and nothing hurting to inject herself is already hard. Handing her a boogeyman makes it impossible. Among Medicare-age adults, 8.01 percent received an anti-osteoporosis medication within a year of a fragility fracture; after hip fracture (which kills roughly one in five within a year), it was 6.87 percent. We are not overtreating this disease.
The table
Open the Prolia (denosumab) label to Table 1. The caption promises adverse reactions occurring in at least 2 percent of patients and more often than on placebo. Now read a row.
From approval in June 2010 until March 2024, that table listed pneumonia: 152 patients on Prolia, 150 on placebo, out of 3,886 and 3,876. Two patients. Fisher exact p = 0.95. Atrial fibrillation sat beside it, 79 versus 77, p = 0.94. In the March 2024 revision, both rows disappeared. The label does not say why; no explanation has been posted and none is required: Adverse-reaction changes are not eligible for the Recent Major Changes box under 21 CFR 201.57(a)(10). A patient who declined Prolia in 2023 over the pneumonia line has no way to learn the entry is gone.
Real physician voices, twice a week
Free, and one click to unsubscribe.
Back pain is still there: 1,347 versus 1,340, p = 0.94. Prolia’s most commonly listed side effect rests on a difference of 7 women out of 7,762. Of the 25 original rows, one reached significance (flatulence, p = 0.0095, which I happily concede as the one thing this table proves) and one more, hypercholesterolemia, landed at p = 0.0501. Across 25 comparisons, chance predicts about 1.25 hits.
A nonsignificant result does not prove two groups identical: The data allow denosumab an extra 22 back-pain cases per 1,000 women, and equally allow it to prevent 20. No label has ever printed “may reduce back pain.” Failing to exclude an effect is not a basis to believe in one. In 7,762 women, 34.7 percent against 34.6 is the background rate of back pain in older women, not a side effect.
The rule, and the caption that gives the game away
An adverse event is something that happened during a trial. An adverse reaction is something there is reason to believe the drug caused. Patients collapse the two into one question (what can this drug do to me?), and most of the damage I see lives in the gap.
The regulators should know the difference: 21 CFR 201.57(c)(7) defines an adverse reaction as “an undesirable effect, reasonably associated with use of a drug” and excludes “all adverse events observed during use of a drug” except those with “some basis to believe there is a causal relationship.” The trial offered no such basis for pneumonia. Perhaps the agency had one elsewhere: mechanism, class effect, another population. If so, the 2024 deletion is harder to explain, not easier: The revision cites nothing. Either there was never a basis, or there was one and it was abandoned in silence.
The caption on Forteo’s (teriparatide) Table 1 ends with eight words: “Adverse Events are Shown Without Attribution of Causality.” Printed by the manufacturer, approved by the agency. Then the Medication Guide, the pharmacy printout, and every consumer website turn it into a list of what the drug does to you. Nobody lied. The document said what it was. It has no defense against being read the way a frightened seventy-year-old will read it.
Evenity (romosozumab) goes further. Its Table 1 requires only that a reaction occur in 2 percent of Evenity-treated women in at least one study: no control-arm comparison at all. It lists eight. In the alendronate trial, all eight were less common on Evenity. Every one. Printed as adverse reactions regardless.
Where this argument stops
These drugs are not free. Zoledronic acid produces an acute-phase reaction in 42.4 percent of patients after the first infusion, against 11.8 percent on placebo: real, enormous, and why patients dread the drip.
Jaw osteonecrosis is the harder case. Rare catastrophes are different: A three-year trial cannot detect them, so absence of significance proves nothing, and the observational data show a dose-response gradient with bisphosphonate duration, the strongest argument for causation short of a trial. The absolute numbers are small but not as small as usually quoted: The American Dental Association gives 0.001 to 0.01 percent for oral bisphosphonates and about 0.1 percent as its highest reliable estimate at osteoporosis doses; the oral surgeons’ society puts denosumab at 0.3 percent. Evenity’s two pivotal trials produced 1 case in 5,621 women. I have watched patients decline treatment for a fracture risk of several percent per year to avoid a jaw risk of, at most, a few in a thousand.
So the principle is not “never warn.” A rare serious event the trial was never powered to see is a legitimate warning. A common symptom the trial measured directly in 7,762 women and found no different is not. That is not an underpowered comparison. It is a precise one that found the same rate in both arms.
What happens downstream
Drugs.com’s Prolia page, updated January 4, 2026, lists among “common Prolia side effects”: lung infection (cough, shortness of breath). Pneumonia left the postmenopausal table in March 2024. The label does still list bronchitis (in Table 2, among glucocorticoid-induced osteoporosis patients), and the consumer page collapses that into a generic lung-infection warning. The rest of its list (back pain, muscle or joint pain, high cholesterol, cold symptoms) is a roll call of null results served as things the drug does to you. Nor is this the first entry the FDA has taken back: Forteo’s osteosarcoma boxed warning stood for eighteen years on a rat study before fifteen years of human surveillance found 3 cases where 4.17 were expected, and nobody wrote to the patients who refused an anabolic agent and fractured in the meantime.
Why
The structural explanation is sufficient. Over-disclosure is free to the discloser. Nobody has ever been sued for having listed pneumonia. The entire cost of a false alarm lands on a patient who is not party to the transaction and will never learn that the line which frightened her was noise.
An older colleague blames the long shadow of Tuskegee and thalidomide: an institutional resolve never to miss anything again, including things that are not there. Unfalsifiable, and I suspect partly right. Either way, what it produces is not safety. It is safetyism: the performance of caution, with the bill sent to someone who cannot see it.
What sustains it is that fear of the FDA puts it beyond scientific criticism, in my opinion. Nobody attacks its arithmetic, and an institution that cannot be told it is wrong does not stop being wrong. It stops finding out.
So I wrote my own handout: A symptom is listed only if it beat placebo. Tymlos (abaloparatide) gets palpitations, 5 percent against 0.4; nausea, 8 against 3; hypercalcemia, 3 against 0.1. Forteo gets one entry: transient hypercalcemia after the dose, 11 percent of women against 2.
That is what a causal standard looks like when you apply it. Sometimes it empties the page. Sometimes it fills it. What it never does is hand a frightened woman twenty-five reasons to say no and leave her to work out which one is real.
Michael Duben is a board-certified endocrinologist in solo private practice in Fairfield, Connecticut, and is affiliated with Bridgeport Hospital. He runs both Endocrinology of Fairfield County, a traditional insurance-based practice, and Restore Health, a cash-based functional endocrinology practice. He completed the Institute for Functional Medicine’s training program.
He did his internal medicine residency at Mount Sinai and his endocrinology fellowship at Montefiore/Einstein, and was a clinical instructor in the Department of Endocrinology at Albert Einstein College of Medicine. He has practiced since 2003, with clinical interests in thyroid disease and thyroid cancer, osteoporosis and bone metabolism, parathyroid disease, pituitary disorders, and obesity and metabolic disease. He has prescribed glucagon-like peptide-1 (GLP-1) receptor agonists since the first one reached the market in 2005, and has lectured to practicing physicians on diabetes and thyroid management.
He designed and built the HIPAA-compliant AI-assisted clinical documentation system he uses in daily practice, without formal training in computer science, along with a range of other tools for his patients and his office. He shares updates on LinkedIn.
