When Implants Fail in BUNCHES
Cluster Failure and Humility
Is there any way to hone in on implant failure, to know the when and why? We know overall failure rates are reported between 2–5% in broad population longitudinal data. With relatively few failures, one might think we could identify the specific factor(s) causing the lack or loss of osseointegration. One would be wrong… at least so far. In fact, the mean failure rates reported may be misleading when applied on a smaller scale, and this in turn may cause us as clinicians to erroneously evaluate risk for any individual patient.
Anecdotally we hear of certain implant patients having “poor protoplasm” (sometimes with colorful alliterative adjectives attached). The idea being although implants in general do very well, particular patients have an unusually high number of failures which are not easily attributable to exogenous causes. Now we qualify this insight — the plural of “anecdote” is not “data,” and anecdotal observations are subject to a slew of cognitive biases. Not the least of which are: negativity bias — wherein we remember unpleasant things more than neutral or pleasant ones of equal (or greater) intensity; and a related availability bias — wherein we place more decision making weight on things which are easier to remember. Knowing these biases and how they color our thinking makes the study at hand all the more useful.
The Research
As with many of the studies we have previously… uh… studied… the authors here are prolific. Dr.’s Chrcanovich, Kitsch, Albrektsson, and Wennerberg have contributed mightily to our collective knowledge of implant failures and successes. Reviewing their library of work is worthwhile and heartily recommended. This paper sheds light on the issue of cluster behavior in implant failure.
The study is retrospective so all related caveats apply [see brief discussion in paragraph three of the 6 implants vs. 4 post]. The population comprised 1406 patients each having received at least 3 implants; there were 8337 total implants and 592 failures. Overall a 7.1% failure rate. But looking closely, things get interesting — the authors defined “cluster behavior” as any patient experiencing 3 or more implant failures. The group exhibiting cluster behavior was composed of 67 patients who received 620 implants and experienced 331 failures. That’s 4.75% of the patients accounting for 56.8% of the total failures of the population. The other 1339 patients received 7717 implants and experienced only 261 failures. That’s 95.25% of patients accounting for 44.1% of failures. Another way of looking at it: 19 of every 20 patients had an implant failure rate around 3.38%, but a bit less than 1 of every 20 had a failure rate around 53.4%. One does not need an advanced degree in statistical analysis to see that something is happening here.
But what? The strongest patient related factors that could be identified in this study were antidepressant use and bruxism. Neither related effect was on the same order of magnitude as the effect seen in the cluster failure group. Other robust studies have looked at these factors:
…but they showed a 60% increased risk of failure with antidepressant use, and at worst something like 280% increased risk for patients diagnosed with bruxism. Still, nothing approaches the 1580% (relative) increased failure rate for the cluster behavior group. It is important to note that none of these contrasts are quite so neat as presented just now, the data reported in studies with different populations, methods, and analytical procedures do not actually stack up to one another in such a straightforward way. The purpose of comparing in this fashion is merely to note that the phenomenon in question here seems at least an order of magnitude different than the most extreme observations of other individual risk factors. This gives us good reason to discount these factors as fully explanatory of the observed failures.
Analysis… or Speculation
So… bruxism and/or antidepressants — no, not those things. Then what? The analysis section falls back on that old villain: “multi-factorial” causes. Always making trouble, those factors… especially when they arrive in multiples! In all seriousness, of course it is possible that the causes of cluster failure are multi-factorial. It’s also possible that there is only one factor causing the failure, and we just haven’t characterized it, or maybe we don’t even realize it exists. “There are more things in heaven and earth… than are dreamt of in your philosophy” and all that. This is the smallest of examples, but it reflects a certain casual lack of intellectual imagination and humility, which is too common in science, and among very well educated people in general. The cold reality is that the data in this study simply do not reveal what has caused these cluster failures at all.
An honest analysis of these data might be: “we have no idea why these cluster failures occur, here is some rank speculation…” As people who treat patients, something within us recoils at the notion of telling a patient “I don’t know.” I suspect this instinct bleeds into scientific writing as well. “We don’t know” is deeply unsatisfying, both to the writer and reader — probably a reason why you so seldom see it in scientific literature. It does not please or flatter either the author or the peer reviewer the way some clever guesses using big words does. Even the suggestion that the analysis is speculation may well offend the sensibilities of the reader (especially the credentialed reader… MD, PhD, lowly DMD… we know who we are). “How dare you call my educated and informed analysis speculation!” we might think to ourselves. Informed and educated in what, though? That which is already known, of course.
But that’s just it — we are versed in that which is known, and may even have a marvelous track record of having predicted new things previously unknown — but here we are, with the data we have, and we are planted firmly in the realm of the unknown. The data give no answer… so any guesses about what is happening are independent of the study. They might just as easily have been made without doing the study at all. They are mostly just re-jiggerings of what we know now. History is a pitiless and indifferent observer of what people (and particularly the medical community) “know” at any given time; see Semmelweis, Ignaz. “How primitive and silly they were!” we may laugh. But are we so blind with hubris that we suppose we are now at the end of learning? That we know how men will judge us 20 or 200 years from now? In whatever the subject, whatever our thoughts may be, let us hope they are not so prideful.
Let’s be humble about what we know, and about our ability to predict the future, but let’s get back to those implants failing. How to assess risk? We have to be aware of these data, AND be aware that we do not yet know how to explain them. Taken together, that means we cannot predict who will exhibit cluster behavior in their implant treatment failures. What to explain to patients preparing for implant procedures? The answer is not prescriptive, and so probably not very edifying: we are the doctors, and we have to make individual judgments. We should not fool ourselves into thinking we might judge who will have these cluster failures in advance. Given the observed existence of the clustering effect, we might suspect that patients who already have had 3 implant failures are perhaps as much as 15 times more likely to have implant failure in the future. Beyond this there is not much more we can say. The best we can do is report the overarching success/failure rates and make an individual choice about how/when/whether to describe the cluster phenomenon to patients. Maybe someday we’ll understand it.
