What a Two-Year Extension Can't Tell You: Reading Long-Term Pain-Drug Data With the Brakes On
Photo by Photo by UK Black Tech on Unsplash on Unsplash
When a pain drug reaches the end of its randomized trial, the story rarely ends. Sponsors often offer participants a long-term extension, usually open-label, in which everyone receives the active drug. In press releases and regulatory summaries, these extensions can sound like a victory lap: two years of data, sustained relief, no new safety signals. As of November 2026, patients and researchers should read them with more caution. An extension is a different study with a different question, and its answers are narrower than they first appear.
The Extension Is a Different Study
In a randomized controlled trial, participants are assigned to drug or control and followed under a fixed protocol. In an open-label extension, the blind comes off, the comparator disappears, and dosing may be adjusted. Visits may become less frequent. Rescue medications may be allowed more freely. The population changes too: only people who completed the randomized phase and chose to continue enter the extension.
That design can answer useful questions. It can show how often serious events accumulate over longer exposure. It can describe persistence, adherence, and reasons for discontinuation. It can generate hypotheses about delayed harms. It cannot, by itself, prove comparative efficacy or establish causal safety in the way a randomized comparison can.
Attrition Is a Result, Not a Footnote
The first number to find in any extension is not the mean pain score. It is the denominator. How many participants started, and how many finished? A two-year extension that begins with 1,200 patients and ends with 700 has lost more than 40 percent of its sample. The remaining patients are not a random subset. They are the ones who tolerated the drug, stayed in the area, and perhaps felt well enough to continue.
That matters most for safety. If patients who develop joint pain stop the drug and leave the extension, the remaining group will look safer than the original cohort. The extension may report a low rate of joint events because the people most likely to have them are no longer being counted. Researchers call this healthy survivor bias, and it is not a small technicality. It can turn a real signal into a reassuring-looking number.
The Comparator That Vanishes
Once the placebo group is gone, pain scores have no internal comparison. A stable pain score over two years may reflect drug effect, natural fluctuation, regression to the mean, continued rescue medication, or simply the selection of patients who were doing well at the end of the randomized phase. Without a concurrent control, the extension cannot tell you how much of that stability is due to the drug.
This is especially important for pain, which waxes and wanes. A patient who enters an extension during a good period may look like a long-term responder. A patient who enters during a flare may look like a failure. The extension design does not fix that problem. It only follows the people who chose to continue.
Joint Safety and the Healthy Survivor
NGF inhibitor programs, including the tanezumab experience, made joint safety a central concern. Joint-related adverse events can be uncommon, delayed, and difficult to adjudicate. An extension can contribute useful information about cumulative incidence, but the interpretation depends on how events were ascertained and who remained at risk. Were joint events adjudicated by a blinded committee? Were imaging studies scheduled, or only ordered after symptoms? Did patients with prior joint replacement enter the extension at all?
A two-year extension can show that a serious joint event rate is low in the people who stayed. It cannot show that the rate would be equally low in the broader population of older adults with osteoarthritis, many of whom would have been excluded or would have discontinued. That distinction should appear in any regulatory or clinical discussion of long-term safety.
A Practical Reading Protocol
When you open an extension paper, follow a fixed sequence. First, locate the extension in the paper and note its start and end dates. Second, count the participants at baseline and at each follow-up. Third, read the discontinuation table by reason, not just the total. Fourth, check whether doses changed and whether rescue medications were allowed. Fifth, confirm that there is no concurrent comparator. Sixth, look for prespecified safety analyses and sensitivity analyses that account for missing data. Seventh, compare the extension population with the randomized population. Eighth, read the confidence intervals, not only the point estimates.
A two-year extension can prove some things. It can show that a drug continues to be taken, that certain rare events accumulate over time, and that some patients report sustained benefit. It cannot prove that the drug is superior to a control it no longer has, or that its long-term safety profile applies to patients who never enrolled. Those limits are not reasons to dismiss extensions. They are reasons to read them as what they are: follow-up studies of volunteers who stayed.