Skip to main content

How Fragile Are Long-Term Implant Outcomes to Loss to Follow-Up? A Methods Review, a Pattern-Mixture Sensitivity Framework, and a Cohort Protocol With Dropout Tracing for Survival, Complications, and Patient-Reported Outcomes

Paper type
Methodological review with theoretical sensitivity analysis and an embedded prospective cohort protocol (no original patient data)
Status
Original Work · Working Paper
Keywords:
dental implants · long-term survival · loss to follow-up · missing not at random · competing risks · complications · supportive care · patient-reported outcomes
Contents
  1. 1Introduction and Clinical Significance
  2. 2Critical Review of the Established Literature
  3. 3Identification of the Specific Research Gap
  4. 4Scientific Objective and Hypothesis
  5. 5Theoretical Foundation: Pattern-Mixture Sensitivity
  6. 6Methodology
  7. 7Evidence Synthesis and Explicitly Labeled Theoretical Results
  8. 8Discussion: Mechanisms and Competing Explanations
  9. 9Implications for Implant Surgery and Restorative Dentistry
  10. 10Technical and Clinical Limitations
  11. 11Proposed Clinical Validation
  12. 12Future Research Directions
  13. 13Evidence-Calibrated Conclusion

Abstract

Background: Ten-year implant survival is reported at about 96 percent in complete-case meta-analysis and lower when missing data are imputed, and estimates at 20 years fall sharply after imputation. Long-term cohorts lose many participants, analyze implants as if independent, treat death as censoring, and rarely report patient-centered outcomes. How much each practice biases each endpoint has not been quantified.

Objective: To review how long-term implant outcomes are estimated, to derive the bias produced by informative loss to follow-up for endpoints of different frequency, and to design a cohort that measures the parameter on which that bias depends.

Methods: Critical review of systematic reviews and methodological literature. A pattern-mixture model was formulated in which implants in patients lost to follow-up have a failure hazard equal to a multiple of that in retained patients. Bias and tipping points were derived in closed form and evaluated using published pooled estimates as inputs. A prospective cohort protocol with intensive tracing of a random sample of dropouts was specified.

Findings: Under the model, the bias in a survival proportion is proportional to the dropout fraction and to the frequency of the event. With a published 10-year implant survival of 96.4 percent, 30 percent dropout, and a hazard among dropouts twice that of completers, the corrected estimate is 95.4 to 95.9 percent; a hazard ratio of roughly 8 to 15 would be required to bring it below 90 percent. With 66.4 percent of patients free of complications at 5 years, the same assumptions change the estimate by 4 to 7 percentage points. Treating death as censoring overstates the cumulative incidence of implant loss modestly. These are calculations from stated assumptions, not observations.

Conclusions: Implant survival is the endpoint least sensitive to the methodological weaknesses of long-term studies and the least informative for patients. Estimates of complication burden, peri-implant disease, and patient-reported outcomes are the ones at risk, and resolving that risk requires tracing those who are lost.

Keywords: dental implants; long-term survival; loss to follow-up; missing not at random; competing risks; complications; supportive care; patient-reported outcomes

1. Introduction and Clinical Significance

Patients who receive dental implants in middle age may live with them for thirty or forty years. Counseling at the time of treatment relies on long-term outcome data, and the figures most often quoted are survival proportions: a recent meta-analysis placed 10-year implant survival at 96.4 percent [1], and an earlier systematic review of studies with at least 10 years of follow-up reported a cumulative mean of 94.6 percent [2]. Such numbers are reassuring and are widely used in consent discussions and in comparisons between treatment options.

They also conceal several things. Survival means only that the implant is still in the mouth; it says nothing about bone loss, infection, prosthetic repair, cost, or the patient's satisfaction. The classic success criteria attempted to add clinical and radiographic conditions [3], later proposals graded implant health [4], and reviews of the literature found the criteria used across studies to be inconsistent [5, 6]. Beyond the choice of endpoint lie questions about how the proportions are estimated. Long-term cohorts lose participants, sometimes most of them. The same meta-analysis that reported 96.4 percent found 93.2 percent in a sensitivity analysis that accounted for missing data, with a prediction interval extending from 76.6 to 100 percent [1]. At 20 years, pooled survival in prospective studies fell from 92 percent to 78 percent after imputation for losses [7].

These sensitivity analyses demonstrate that the question matters; they cannot settle it, because any imputation rests on an assumption about people who were not seen. This paper approaches the problem from the other direction. It asks how large the bias could be, for which endpoints, under what assumptions, and what would have to be measured to replace assumption with evidence.

2. Critical Review of the Established Literature

Survival of implants

Systematic reviews agree on high medium-term implant survival. For implants supporting single crowns, pooled survival was 97.2 percent at 5 years and 95.2 percent at 10 years [8]; for implants supporting fixed dental prostheses, 95.6 percent and 93.1 percent [9]. A comparison of older and newer publications found higher survival and fewer complications in the more recent studies [10]. Large single-center and practice-based series report similar or better figures: a retrospective study of 511 implants with a sandblasted and acid-etched surface found 10-year survival close to 99 percent [11], and a private-practice cohort of more than four thousand implants reported cumulative implant-level survival above 98 percent at 7 years, with patient-level survival somewhat lower [12]. Data from a single specialist clinic spanning several decades of implant operations provide some of the longest series available [13].

Population-based data are less favorable than specialist series. In a random sample from a national register, 7.6 percent of patients had lost at least one implant after about nine years, counting early and late losses [14]. The difference between effectiveness in general practice and efficacy in selected centers is a form of selection bias in the literature as a whole.

Survival of prostheses and complications

Prostheses fail more often than implants. Implant-supported single crowns survived in 96.3 percent of cases at 5 years and 89.4 percent at 10 years [8]. Fixed dental prostheses on implants survived in 95.4 percent at 5 years and 80.1 percent at 10 years overall, and in 93.9 percent at 10 years when the analysis was confined to metal-ceramic prostheses; only 66.4 percent of patients were free of any complication after 5 years, with veneer fracture, peri-implantitis and soft-tissue complications, loss of access-hole restoration, and screw or abutment loosening the most frequent events [9]. Systematic reviews of zirconia-ceramic and metal-ceramic implant restorations add material-specific complication profiles [15, 16], and for fixed complete-arch rehabilitation the cumulative burden of biological and technical complications over 5 to 10 years is high [17].

Biological complications accumulate with time. Weighted mean prevalences of peri-implant mucositis and peri-implantitis in the epidemiological literature are about 43 and 22 percent of patients [18], with case definitions specified by the 2017 World Workshop [19]. Peri-implantitis has a further ascertainment problem. It is detected only at examination, so that its onset is interval-censored, and cross-sectional prevalence among those attending at a given anniversary is commonly reported in place of incidence, with case definitions and reporting quality that vary between studies [20]. Prevalence among attenders is doubly exposed to selective attendance: through who remains in the cohort, and through who presents for the examination at which disease is assessed.

Maintenance

Supportive care modifies long-term outcomes. A meta-analysis found that maintenance therapy reduced the occurrence of peri-implant diseases [21]; in patients with mucositis, the incidence of peri-implantitis over five years was considerably higher among those without preventive maintenance [22]; and in a 10-year prospective cohort of periodontally compromised and healthy patients, outcomes were poorer in those with a history of periodontitis, particularly when they did not adhere to supportive periodontal therapy [23]. Adherence to maintenance and retention in a follow-up study are closely related behaviors. Patients who stop attending maintenance are the ones who leave the cohort, which is the mechanism by which loss to follow-up could be informative.

Patient-reported outcomes

Patient-reported outcome measures capture what survival does not. Reviews found that implant studies used patient-reported outcomes inconsistently, frequently with unvalidated instruments [24, 25], and a consensus report called for standardized, validated measures [26]. Validated generic instruments exist, of which the short-form Oral Health Impact Profile is the most widely used [27]. An international consensus has since defined a core outcome set for implant dentistry that includes patient-reported domains alongside clinical ones [28]. Long-term patient-reported data remain rare, and they are subject to the same attrition as clinical data, with the added complication that a questionnaire can be missing even when the patient attends.

Older patients and competing mortality

Implant survival in older adults is high in pooled analyses [29], and a systematic review of patients aged 75 and over likewise reported high survival at 1 and 5 years [30]; the 10-year meta-analysis nevertheless found lower survival in those aged 65 and over in its sensitivity analysis [1]. In older cohorts followed for 10 to 20 years, a substantial fraction die. Standard Kaplan-Meier analysis censors them at death, which estimates the probability of implant loss in a hypothetical world without death and overstates the probability that a real patient will experience implant loss [31].

Statistical practice

Three statistical issues recur. The first is clustering. Implants in one patient share that patient's risk factors, and analyses that treat them as independent understate uncertainty. The effect on point estimates of survival is small when failure is rare, as shown in a comparison of three analytic strategies in a large cohort, but the standard errors differ, most in high-risk subgroups [32]; the consequence of implant interdependency for reported success rates was demonstrated on multicenter trial data [33].

The second is missing data. Methods for incomplete data distinguish data missing completely at random, missing at random given observed variables, and missing not at random, where missingness depends on the unobserved outcome itself [34]. Standard survival analysis assumes that censoring is non-informative. Multiple imputation relaxes the assumption to missing at random and has well-known pitfalls [35]. Neither addresses data missing not at random, for which the recommended approach is sensitivity analysis: pattern-mixture models in which the departure from missing at random is indexed by a parameter varied over a plausible range, and tipping-point analysis that reports the parameter value at which the conclusion would change [36, 37]. A review of randomized trials in major journals found that plausible assumptions about participants lost to follow-up could change the conclusions of a sizeable proportion of trials [38].

The third is the reporting of observational studies and of studies using routinely collected data, for which guidelines exist [39, 40]. Implant registers have been proposed as a way to obtain unselected long-term data [41].

3. Identification of the Specific Research Gap

What is established: implant survival at 10 years is high in followed patients; prosthetic and biological complications are common; maintenance improves outcomes; long-term studies lose many participants; and sensitivity analyses that impute outcomes for those lost produce lower estimates.

What is uncertain is the direction and size of the bias from loss to follow-up for each endpoint. Two opposite mechanisms are plausible. Patients who stop attending may be those who neglect maintenance and whose implants fare worse. Alternatively, patients with problems return for treatment while those without problems see no reason to attend, in which case those lost fare better than those seen. The literature contains no direct evidence on which predominates in implant cohorts.

The methods are inadequate because imputation-based sensitivity analyses choose an assumption, commonly a pessimistic one, and report a single alternative estimate, which conveys that the truth lies between two numbers when in fact it depends on an unmeasured quantity. The quantity in question, the relative hazard among those lost, has not been estimated in any implant study we identified. In addition, the endpoints analyzed are those least sensitive to the problem. Implant loss is rare, so that even strongly informative dropout moves the estimate little, whereas the frequent endpoints that matter to patients are seldom subjected to any sensitivity analysis.

4. Scientific Objective and Hypothesis

The objective is to quantify the sensitivity of long-term implant outcome estimates to informative loss to follow-up, competing mortality, and clustering, and to design a study that estimates the governing sensitivity parameter empirically.

Hypothesis H18 has two parts. H18a: implant-level survival at 10 years is robust to informative dropout, in the sense that the estimate corrected using outcomes ascertained in a traced random sample of dropouts differs from the complete-case estimate by less than 2 percentage points. H18b: patient-level complication-free survival at 10 years is not robust, in the sense that the corrected estimate differs from the complete-case estimate by 5 percentage points or more, in either direction.

H18a is falsified if the corrected implant survival differs from the complete-case value by 2 points or more with a confidence interval for the difference excluding values below 2. H18b is falsified if the corrected complication-free survival differs by less than 5 points with a confidence interval excluding 5. The thresholds are stated as judgments of clinical relevance: a 2-point difference in implant survival would not change counseling, while a 5-point difference in the probability of remaining free of complications would.

The hypothesis was framed after the theoretical analysis in Section 5 and is consistent with it by construction; its empirical content lies in the unknown value of the dropout hazard ratio, which could be large enough to refute H18a or close enough to 1 to refute H18b.

5. Theoretical Foundation: Pattern-Mixture Sensitivity

Model

Let Sobs(T) be the proportion event-free at time T estimated from participants who remained under observation, and let d be the fraction lost to follow-up before T. Assume a constant event hazard λ in retained participants, so that

S_(obs)(T) = e^(-λ T) (1)

and assume that after leaving the study a participant's hazard is δλ, where δ is the dropout hazard ratio. A value of δ equal to 1 corresponds to non-informative loss, δ above 1 to worse outcomes among those lost, and δ below 1 to better outcomes. The true proportion event-free in the whole cohort is the mixture

S_(true)(T) = (1 - d) S_(obs)(T) + d S_(drop)(T) (2)

The term Sdrop depends on when participants leave. Two bounding cases are considered. If all losses occur at baseline, the most extreme case,

S_drop^early(T) = S_(obs)(T)^δ (3)

If losses occur uniformly over the interval from 0 to T, a participant leaving at time u is exposed to hazard λ until u and δλ thereafter, and averaging over u gives

S_drop^unif(T) = (e^(-λ T) - e^(-δ λ T))/((δ - 1) λ T) (4)

for δ not equal to 1. Real cohorts lie between the two.

Bias and its dependence on event frequency

In the early-loss case the bias of the complete-case estimate is

S_(obs) - S_(true) = d S_(obs) (1 - S_obs^δ - 1) (5)

For a rare event, Sobs is close to 1 and the bias is approximately d(δ − 1)(1 − Sobs): the dropout fraction, times the excess hazard ratio, times the event probability. The bias is therefore small whenever the event is rare, almost regardless of how informative the dropout is. For a common event the factor in parentheses is no longer small and the bias can be many percentage points. This is the central theoretical point of the paper: the same loss to follow-up that leaves implant survival nearly unbiased can seriously bias the estimated frequency of complications.

Tipping points

The tipping point δ* for a threshold Scrit is the value of δ at which Strue equals Scrit. In the early-loss case it has the closed form

δ^* = (ln[(S_(crit) - (1 - d) S_(obs)) / d])/(ln S_(obs)) (6)

provided the argument of the logarithm is positive; otherwise no value of δ reaches the threshold. Reporting δ* in place of, or alongside, an imputed estimate tells the reader how extreme the departure from non-informative loss would have to be to change a conclusion, and leaves the judgment of plausibility explicit.

Competing mortality

If death occurs at constant hazard μ and removes the participant from risk, the cumulative incidence of implant loss by time T is

F(T) = λ/(λ + μ) (1 - e^(-(λ + μ) T)) (7)

which is smaller than the Kaplan-Meier complement 1 − exp(−λT) obtained by censoring deaths. Regression on the cumulative incidence scale uses the subdistribution hazard [42]. The Kaplan-Meier quantity answers an etiological question about implant durability; the cumulative incidence answers the patient's question about what is likely to happen in a real lifetime [31].

Unit of analysis

With m implants per patient and implant-level survival S, the probability that a patient loses at least one implant would be 1 − Sm if implants failed independently. Because failures cluster within patients, the true patient-level proportion is lower than this and the proportion of patients with multiple failures is higher. Implant-level and patient-level proportions are different estimands, both legitimate, and a report should state which it gives and how clustering was handled.

Continuous patient-reported outcomes

For a continuous outcome such as a quality-of-life score with mean μobs among responders, if a fraction m of scores is missing and the mean among those missing differs from that among responders by Δ, the mean in the whole cohort is

μ_(true) = μ_(obs) + m Δ (8)

Here there is no rarity term to limit the bias, which is simply the product of the missing fraction and the shift. Expressed in units of the between-patient standard deviation, a shift of half a standard deviation with 30 percent of scores missing biases the mean by 0.15 standard deviations, and a shift of one standard deviation with half of the scores missing biases it by 0.5. Since differences of about half a standard deviation have been proposed as a general benchmark for a minimally important change in health-related quality of life [43], missingness of the degree usual in long-term implant studies is capable of producing or concealing a difference of clinical importance in a patient-reported outcome.

6. Methodology

Review methods

The review is a critical narrative synthesis centered on published systematic reviews for outcome frequencies and on methodological literature for analytic issues. It is not a systematic review.

Theoretical calculations

Equations 2 to 7 were evaluated with published pooled estimates as inputs: 10-year implant survival of 96.4 percent [1] and a 5-year proportion of patients free of complications of 66.4 percent among those with implant-supported fixed dental prostheses [9]. Dropout fractions of 0.2, 0.3, and 0.5 and hazard ratios of 0.5, 2, and 4 were chosen to span optimistic and pessimistic scenarios; they are assumptions and not estimates. The calculations are deterministic and reproducible from the equations; the script used is retained with the manuscript files. Results are presented in Section 7.2 and are labeled as theoretical.

Proposed cohort

Design. A prospective, multicenter cohort of consecutive adults receiving implant-supported restorations in specialist and general practice settings, reported according to STROBE [39], with linkage to routinely collected data where available and reporting of the linked components according to RECORD [40].

Follow-up. Annual recall for clinical examination, with standardized radiographs at prosthesis delivery, 1, 5, and 10 years. Supportive care is delivered according to local practice and attendance is recorded as an exposure.

Outcomes. Outcomes follow the international core outcome set [28]: implant loss; prosthesis loss or remake; technical complications by type; peri-implant mucositis and peri-implantitis by the 2017 definitions [19]; marginal bone level; and patient-reported outcomes by a validated oral health-related quality of life instrument [27] and a treatment-specific satisfaction measure. Death and its date are obtained from registries.

Dropout tracing. The distinguishing feature is a double-sampling component. Every participant who misses two consecutive annual recalls is classified as lost. From those lost, a random sample, stratified by center and time since treatment, is selected for intensive tracing by telephone, contact with the current dentist, record linkage, and, where the participant agrees, a home or local examination with a portable radiograph. The outcomes ascertained in the traced sample are weighted by the inverse of the sampling fraction to represent all those lost. The approach was developed for cohort studies in which loss to follow-up is substantial and self-selected, where it removed large biases in survival estimates [44, 45].

Estimands. For each endpoint the cohort reports the complete-case estimate, the tracing-corrected estimate, the estimated dropout hazard ratio δ with its confidence interval, and the tipping point δ* for a prespecified threshold. Implant-level analyses use methods for clustered survival data [32]; patient-level analyses define the event as the first occurrence in any implant or prosthesis. Cumulative incidence functions with death as a competing event are reported alongside Kaplan-Meier estimates [31, 42].

Table 1. Estimands and analyses in the proposed cohort.
EndpointUnitAnalyses reported
Implant lossImplant and patientKaplan-Meier with clustering; cumulative incidence with death competing; tracing-corrected estimate; δ and δ*
Prosthesis failure or remakeProsthesisAs above
Any technical or biological complicationPatientTime to first event; complication-free proportion; tracing-corrected estimate; δ and δ*
Peri-implantitisImplant and patientPrevalence at examination; interval-censored incidence; tracing-corrected estimate
Oral health-related quality of lifePatientMixed models for repeated measures; pattern-mixture sensitivity analysis with tracing data
Maintenance attendancePatientTime-varying exposure; association with outcomes and with dropout

Sample size

The precision of the estimated dropout hazard ratio depends on the number of events among traced dropouts. For the complication endpoint, if one third of patients experience a complication within 5 years, about 240 traced dropouts would be expected to yield of the order of 80 events, sufficient to estimate δ with a confidence interval whose upper and lower limits differ by a factor of about 1.5 to 2. For implant loss, with an event probability of a few percent, the same sample would yield about 10 events, and δ would be estimated imprecisely; this is acceptable because, by Equation 5, the corrected survival estimate is insensitive to δ over a wide range. If 30 percent of a cohort of 4,000 patients were lost by 10 years and one quarter of them were sampled for tracing with 80 percent success, about 240 would be traced. These figures are planning assumptions.

7. Evidence Synthesis and Explicitly Labeled Theoretical Results

Synthesis of published estimates

Published estimates support three statements with reasonable confidence: implants that are followed survive in about 95 percent of cases at 10 years; prostheses on them need repair or replacement much more often than the implants are lost; and a large minority of patients experience some complication within 5 years. The estimates derive mostly from specialist and university settings, and register-based data suggest less favorable results in general practice [14].

Theoretical results from the pattern-mixture model

The results in Table 2 are calculations from Equations 2 to 6 with the inputs and assumptions stated in Section 6.2. They are not empirical findings.

Table 2. Theoretical results: true event-free proportion (percent) implied by a published complete-case estimate under assumed dropout fraction d and dropout hazard ratio δ. Ranges give the uniform-loss and early-loss bounds.
Endpoint and published complete-case valuedδ = 0.5δ = 2δ = 4
Implant survival, 10 years, 96.4 [1]0.296.6 to 96.895.7 to 96.194.4 to 95.4
0.396.7 to 96.995.4 to 95.993.4 to 94.9
0.596.8 to 97.394.7 to 95.591.4 to 93.8
Patients free of complications, 5 years, 66.4 [9]0.267.9 to 69.461.9 to 64.057.0 to 60.8
0.368.6 to 70.959.7 to 62.852.3 to 57.9
0.570.0 to 73.955.2 to 60.442.9 to 52.3

Three observations follow. First, for implant survival, a dropout hazard ratio of 2 with 30 percent loss changes the estimate by about 0.5 to 1 percentage point. The tipping points for a 90 percent threshold are large: about 8 under the early-loss bound and about 15 under uniform loss with 30 percent dropout, and about 5 and 9 with 50 percent dropout. Unless those lost fare many times worse than those followed, 10-year implant survival remains above 90 percent.

Second, for the complication-free proportion the same assumptions shift the estimate by 4 to 7 points at 30 percent loss, and the tipping point for a 5-point change is a hazard ratio between about 1.7 and 2.5. A modest and entirely plausible degree of informative loss would therefore change this estimate by a clinically relevant amount, and in either direction, since a hazard ratio of 0.5 raises it by 2 to 5 points.

Third, for competing mortality, with 10-year implant survival of 96.4 percent the Kaplan-Meier complement is 3.6 percent, while the cumulative incidence from Equation 7 is 3.4, 3.0, and 2.6 percent when 10, 30, and 50 percent of patients die within the period. The overstatement is small in absolute terms for a rare event and would be proportionally the same, and absolutely larger, for common events in elderly cohorts.

A further calculation concerns the unit of analysis. With implant survival of 96.4 percent, independence would imply that 7.1, 13.6, and 19.7 percent of patients with two, four, and six implants lose at least one. The register-based observation that 7.6 percent of patients lost at least one implant over about nine years [14] is of this order; the comparison is only indicative, since the populations and implant numbers differ.

8. Discussion: Mechanisms and Competing Explanations

The analysis reverses a common emphasis. Discussion of loss to follow-up in implant dentistry has concentrated on survival, where pessimistic imputation yields alarming-looking reductions. The pattern-mixture model shows that large reductions in survival require dropout hazard ratios that are hard to credit as an average over all those lost, although they could be true of subgroups. The published 20-year estimate that fell from 92 to 78 percent after imputation [7] is not in conflict with this: at 20 years events are less rare, losses are heavier, and the imputation model may embody a strong assumption. The same reasoning scales to longer horizons: with observed survival near 90 percent and half of the original cohort unobserved, Equation 5 gives a bias of about 4 to 5 percentage points for a dropout hazard ratio of 2 and about 12 points for a ratio of 4 under the early-loss bound, so that at 20 years survival itself becomes a sensitive endpoint. The appropriate response to such a result is to ask what assumption about the unobserved produced it and whether that assumption is believable, which a tipping-point presentation makes transparent.

For complications, the analysis points the other way. Estimates such as two thirds of patients free of complications at 5 years come from patients who were examined, as the source review required [9]. If those who stopped attending had more untreated complications, the true burden is higher; if those who attended were disproportionately those with problems, it is lower. Both mechanisms operate in dental practice, and their balance probably differs by setting, payment system, and type of complication. A fractured veneer or a loose screw brings the patient back; early peri-implantitis is painless and does not. The direction of bias may therefore differ between technical and biological complications, with technical complications over-represented and biological complications under-represented in data from attending patients. This is a hypothesis that the tracing design could test directly by estimating δ separately for each.

There are alternative views of the whole question. One is that loss to follow-up in implant studies is mostly administrative, arising from relocation and practice changes unrelated to oral health, and that δ is close to 1. If so, complete-case estimates are sound and H18b fails. Another is that the more serious bias lies upstream, in who is selected for implant treatment and whose results are published, and that no amount of care over follow-up within a specialist cohort will produce estimates applicable to general practice; register-based sampling addresses that concern better than tracing [14, 41]. The two are complementary and a complete program would include both.

The model has evident simplifications. Constant hazards do not describe implant failure, which has an early peak and a later slow rise, nor peri-implantitis, which accelerates. A single δ for all those lost ignores heterogeneity; a mixture in which most have δ near 1 and a minority have very high δ would produce the same average bias with different implications for individuals. The early and uniform cases bound simple timing patterns but not a pattern in which losses concentrate just before events. These refinements would change numerical values in Table 2 by amounts smaller than the difference between its columns and would not change the dependence of bias on event frequency, which follows from the mixture structure alone.

Patient-reported outcomes deserve separate comment. Their missingness is plausibly related to their value, since dissatisfied patients may disengage or, conversely, be more motivated to respond. No implant study we identified reported a sensitivity analysis for missing patient-reported outcomes. Because these are continuous measures with no rarity term to protect them (Equation 8), they are likely to be at least as sensitive to informative missingness as complication proportions.

Finally, the result has a constructive side. If implant survival is robust, then effort spent refining survival estimates has diminishing returns, and the same follow-up resources would yield more information if directed at complications, disease, maintenance, and patient-reported outcomes, with tracing to protect those estimates.

9. Implications for Implant Surgery and Restorative Dentistry

For counseling, the analysis supports continued use of published 10-year implant survival figures as reasonable estimates for patients who resemble those studied and who attend follow-up, with the acknowledgment that results in general practice may be somewhat lower. It does not support equal confidence in published complication-free proportions, which should be presented as approximate and possibly optimistic for biological complications. Patients should be told that needing some repair or treatment within 5 to 10 years is common, that this is distinct from losing the implant, and that regular supportive care is the factor under their control with the best evidence.

For practices, recording maintenance attendance and making an effort to learn the outcome of patients who stop attending has scientific as well as clinical value. For authors and reviewers, long-term reports should state the number lost at each interval and the reasons where known, give implant-level and patient-level results with clustering handled explicitly, report cumulative incidence when deaths are frequent, include patient-reported outcomes with their missingness, and replace single imputed estimates with tipping-point analyses.

10. Technical and Clinical Limitations

The theoretical results rest on a simple model with constant hazards and a single sensitivity parameter, and on published pooled estimates taken as fixed inputs although they carry their own uncertainty and heterogeneity. The dropout fractions and hazard ratios examined are assumptions. The review is not systematic. The proposed tracing design is demanding: it depends on consent to be contacted after disengagement, on registry access that varies between jurisdictions, and on examinations outside the clinic that may be less accurate than those within it; traced participants who can be found and agree may differ from those who cannot, leaving a residual, smaller missing-data problem. Ten-year follow-up means that results would describe implants and prosthetic materials already superseded. And the complication-free proportion is a coarse summary that weights a chipped veneer equally with advanced peri-implantitis; severity-weighted measures would be preferable but are not standardized.

11. Proposed Clinical Validation

Validation of the framework can begin without new patients. Existing long-term cohorts that recorded reasons for loss, or that can be linked to insurance or national registers, could estimate δ for implant loss and for reintervention retrospectively. Authors of published long-term studies could be asked to report tipping points for their principal estimates from data they already hold. A simulation study is a further inexpensive step: synthetic cohorts with known hazards, within-patient clustering, time-varying dropout, and mixtures of dropout hazard ratios can be generated to check how well the two-bound approximation and the tracing-weighted estimator recover the truth under conditions more realistic than those assumed in Section 5, and to choose the tracing sampling fraction. The proposed prospective cohort then provides the definitive test, with an internal pilot of the tracing procedure in the first two years to establish the achievable tracing success and cost per traced participant. H18a and H18b are evaluated at 10 years against the criteria in Section 4, with an interim analysis at 5 years for the complication endpoint.

12. Future Research Directions

Several extensions follow. Severity-weighted and cost-weighted complication measures would align outcome reporting with patient and payer perspectives. Joint models of maintenance attendance, dropout, and outcome could estimate how much of the association between attendance and outcome is causal. Register-based designs with random sampling from the treated population, as used in Sweden, could be replicated in other health systems to address selection into the literature. Minimal important differences for patient-reported measures specific to implant therapy need to be established before long-term changes can be interpreted. For the growing population of older adults with implants, outcomes under declining dexterity and increasing dependency, including the feasibility of maintenance in institutional care, are scarcely documented.

13. Evidence-Calibrated Conclusion

Published evidence of moderate quality indicates that about 95 percent of implants in followed patients survive 10 years and that prosthetic and biological complications are frequent. A pattern-mixture analysis, which is theoretical and depends on stated assumptions, indicates that the survival estimate is robust to informative loss to follow-up unless those lost fare several times worse than those followed, while estimates of complication-free survival are sensitive to modest departures from non-informative loss in either direction. The relative outcome of patients lost to follow-up has not been measured in implant dentistry. A cohort with dropout tracing would measure it and would show whether the confidence placed in long-term complication and patient-reported outcome data is warranted. Until then, survival figures can be quoted with reasonable assurance, and figures for everything else with less.

Cite this paper

Jain S, Gupta A. How Fragile Are Long-Term Implant Outcomes to Loss to Follow-Up? A Methods Review, a Pattern-Mixture Sensitivity Framework, and a Cohort Protocol With Dropout Tracing for Survival, Complications, and Patient-Reported Outcomes. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/dropout-sensitivity-of-long-term-implant-outcomes

References

  1. Howe MS, Keys W, Richards D. Long-term (10-year) dental implant survival: A systematic review and sensitivity meta-analysis. J Dent. 2019;84:9-21. doi:10.1016/j.jdent.2019.03.008 (opens in a new tab) PMID: 30904559 (opens in a new tab).
  2. Moraschini V, Poubel LA, Ferreira VF, Barboza Edos S. Evaluation of survival and success rates of dental implants reported in longitudinal studies with a follow-up period of at least 10 years: a systematic review. Int J Oral Maxillofac Surg. 2015;44(3):377-88. doi:10.1016/j.ijom.2014.10.023 (opens in a new tab) PMID: 25467739 (opens in a new tab).
  3. Albrektsson T, Zarb G, Worthington P, Eriksson AR. The long-term efficacy of currently used dental implants: a review and proposed criteria of success. Int J Oral Maxillofac Implants. 1986;1(1):11-25. PMID: 3527955 (opens in a new tab).
  4. Misch CE, Perel ML, Wang HL, Sammartino G, Galindo-Moreno P, Trisi P, et al. Implant success, survival, and failure: the International Congress of Oral Implantologists (ICOI) Pisa Consensus Conference. Implant Dent. 2008;17(1):5-15. doi:10.1097/ID.0b013e3181676059 (opens in a new tab) PMID: 18332753 (opens in a new tab).
  5. Papaspyridakos P, Chen CJ, Singh M, Weber HP, Gallucci GO. Success criteria in implant dentistry: a systematic review. J Dent Res. 2012;91(3):242-8. doi:10.1177/0022034511431252 (opens in a new tab) PMID: 22157097 (opens in a new tab).
  6. Needleman I, Chin S, O'Brien T, Petrie A, Donos N. Systematic review of outcome measurements and reference group(s) to evaluate and compare implant success and failure. J Clin Periodontol. 2012;39 Suppl 12:122-32. doi:10.1111/j.1600-051X.2011.01836.x (opens in a new tab) PMID: 22533952 (opens in a new tab).
  7. Kupka JR, König J, Al-Nawas B, Sagheb K, Schiegnitz E. How far can we go? A 20-year meta-analysis of dental implant survival rates. Clin Oral Investig. 2024;28(10):541. doi:10.1007/s00784-024-05929-3 (opens in a new tab) PMID: 39305362 (opens in a new tab).
  8. Jung RE, Zembic A, Pjetursson BE, Zwahlen M, Thoma DS. Systematic review of the survival rate and the incidence of biological, technical, and aesthetic complications of single crowns on implants reported in longitudinal studies with a mean follow-up of 5 years. Clin Oral Implants Res. 2012;23 Suppl 6:2-21. doi:10.1111/j.1600-0501.2012.02547.x (opens in a new tab) PMID: 23062124 (opens in a new tab).
  9. Pjetursson BE, Thoma D, Jung R, Zwahlen M, Zembic A. A systematic review of the survival and complication rates of implant-supported fixed dental prostheses (FDPs) after a mean observation period of at least 5 years. Clin Oral Implants Res. 2012;23 Suppl 6:22-38. doi:10.1111/j.1600-0501.2012.02546.x (opens in a new tab) PMID: 23062125 (opens in a new tab).
  10. Pjetursson BE, Asgeirsson AG, Zwahlen M, Sailer I. Improvements in implant dentistry over the last decade: comparison of survival and complication rates in older and newer publications. Int J Oral Maxillofac Implants. 2014;29 Suppl:308-24. doi:10.11607/jomi.2014suppl.g5.2 (opens in a new tab) PMID: 24660206 (opens in a new tab).
  11. Buser D, Janner SF, Wittneben JG, Brägger U, Ramseier CA, Salvi GE. 10-year survival and success rates of 511 titanium implants with a sandblasted and acid-etched surface: a retrospective study in 303 partially edentulous patients. Clin Implant Dent Relat Res. 2012;14(6):839-51. doi:10.1111/j.1708-8208.2012.00456.x (opens in a new tab) PMID: 22897683 (opens in a new tab).
  12. French D, Larjava H, Ofec R. Retrospective cohort study of 4591 Straumann implants in private practice setting, with up to 10-year follow-up. Part 1: multivariate survival analysis. Clin Oral Implants Res. 2015;26(11):1345-54. doi:10.1111/clr.12463 (opens in a new tab) PMID: 25134415 (opens in a new tab).
  13. Jemt T, Olsson M, Franke Stenport V. Incidence of First Implant Failure: A Retroprospective Study of 27 Years of Implant Operations at One Specialist Clinic. Clin Implant Dent Relat Res. 2015;17 Suppl 2:e501-10. doi:10.1111/cid.12277 (opens in a new tab) PMID: 25536273 (opens in a new tab).
  14. Derks J, Håkansson J, Wennström JL, Tomasi C, Larsson M, Berglundh T. Effectiveness of implant therapy analyzed in a Swedish population: early and late implant loss. J Dent Res. 2015;94(3 Suppl):44S-51S. doi:10.1177/0022034514563077 (opens in a new tab) PMID: 25503901 (opens in a new tab).
  15. Pjetursson BE, Valente NA, Strasding M, Zwahlen M, Liu S, Sailer I. A systematic review of the survival and complication rates of zirconia-ceramic and metal-ceramic single crowns. Clin Oral Implants Res. 2018;29 Suppl 16:199-214. doi:10.1111/clr.13306 (opens in a new tab) PMID: 30328190 (opens in a new tab).
  16. Sailer I, Strasding M, Valente NA, Zwahlen M, Liu S, Pjetursson BE. A systematic review of the survival and complication rates of zirconia-ceramic and metal-ceramic multiple-unit fixed dental prostheses. Clin Oral Implants Res. 2018;29 Suppl 16:184-198. doi:10.1111/clr.13277 (opens in a new tab) PMID: 30328185 (opens in a new tab).
  17. Papaspyridakos P, Chen CJ, Chuang SK, Weber HP, Gallucci GO. A systematic review of biologic and technical complications with fixed implant rehabilitations for edentulous patients. Int J Oral Maxillofac Implants. 2012;27(1):102-10. PMID: 22299086 (opens in a new tab).
  18. Derks J, Tomasi C. Peri-implant health and disease. A systematic review of current epidemiology. J Clin Periodontol. 2015;42 Suppl 16:S158-71. doi:10.1111/jcpe.12334 (opens in a new tab) PMID: 25495683 (opens in a new tab).
  19. Berglundh T, Armitage G, Araujo MG, Avila-Ortiz G, Blanco J, Camargo PM, et al. Peri-implant diseases and conditions: Consensus report of workgroup 4 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Periodontol. 2018;89 Suppl 1:S313-S318. doi:10.1002/JPER.17-0739 (opens in a new tab) PMID: 29926955 (opens in a new tab).
  20. Tomasi C, Derks J. Clinical research of peri-implant diseases--quality of reporting, case definitions and methods to study incidence, prevalence and risk factors of peri-implant diseases. J Clin Periodontol. 2012;39 Suppl 12:207-23. doi:10.1111/j.1600-051X.2011.01831.x (opens in a new tab) PMID: 22533958 (opens in a new tab).
  21. Monje A, Aranda L, Diaz KT, Alarcón MA, Bagramian RA, Wang HL, et al. Impact of Maintenance Therapy for the Prevention of Peri-implant Diseases: A Systematic Review and Meta-analysis. J Dent Res. 2016;95(4):372-9. doi:10.1177/0022034515622432 (opens in a new tab) PMID: 26701350 (opens in a new tab).
  22. Costa FO, Takenaka-Martinez S, Cota LO, Ferreira SD, Silva GL, Costa JE. Peri-implant disease in subjects with and without preventive maintenance: a 5-year follow-up. J Clin Periodontol. 2012;39(2):173-81. doi:10.1111/j.1600-051X.2011.01819.x (opens in a new tab) PMID: 22111654 (opens in a new tab).
  23. Roccuzzo M, Bonino L, Dalmasso P, Aglietta M. Long-term results of a three arms prospective cohort study on implants in periodontally compromised patients: 10-year data around sandblasted and acid-etched (SLA) surface. Clin Oral Implants Res. 2014;25(10):1105-12. doi:10.1111/clr.12227 (opens in a new tab) PMID: 23865554 (opens in a new tab).
  24. McGrath C, Lam O, Lang N. An evidence-based review of patient-reported outcome measures in dental implant research among dentate subjects. J Clin Periodontol. 2012;39 Suppl 12:193-201. doi:10.1111/j.1600-051X.2011.01841.x (opens in a new tab) PMID: 22533956 (opens in a new tab).
  25. De Bruyn H, Raes S, Matthys C, Cosyn J. The current use of patient-centered/reported outcomes in implant dentistry: a systematic review. Clin Oral Implants Res. 2015;26 Suppl 11:45-56. doi:10.1111/clr.12634 (opens in a new tab) PMID: 26385620 (opens in a new tab).
  26. Feine J, Abou-Ayash S, Al Mardini M, de Santana RB, Bjelke-Holtermann T, Bornstein MM, et al. Group 3 ITI Consensus Report: Patient-reported outcome measures associated with implant dentistry. Clin Oral Implants Res. 2018;29 Suppl 16:270-275. doi:10.1111/clr.13299 (opens in a new tab) PMID: 30328187 (opens in a new tab).
  27. Slade GD. Derivation and validation of a short-form oral health impact profile. Community Dent Oral Epidemiol. 1997;25(4):284-90. doi:10.1111/j.1600-0528.1997.tb00941.x (opens in a new tab) PMID: 9332805 (opens in a new tab).
  28. Tonetti MS, Sanz M, Avila-Ortiz G, Berglundh T, Cairo F, Derks J, et al. Relevant domains, core outcome sets and measurements for implant dentistry clinical trials: The Implant Dentistry Core Outcome Set and Measurement (ID-COSM) international consensus report. J Clin Periodontol. 2023;50 Suppl 25:5-21. doi:10.1111/jcpe.13808 (opens in a new tab) PMID: 37143289 (opens in a new tab).
  29. Srinivasan M, Meyer S, Mombelli A, Müller F. Dental implants in the elderly population: a systematic review and meta-analysis. Clin Oral Implants Res. 2017;28(8):920-930. doi:10.1111/clr.12898 (opens in a new tab) PMID: 27273468 (opens in a new tab).
  30. Schimmel M, Srinivasan M, McKenna G, Müller F. Effect of advanced age and/or systemic medical conditions on dental implant survival: A systematic review and meta-analysis. Clin Oral Implants Res. 2018;29 Suppl 16:311-330. doi:10.1111/clr.13288 (opens in a new tab) PMID: 30328186 (opens in a new tab).
  31. Austin PC, Lee DS, Fine JP. Introduction to the Analysis of Survival Data in the Presence of Competing Risks. Circulation. 2016;133(6):601-9. doi:10.1161/CIRCULATIONAHA.115.017719 (opens in a new tab) PMID: 26858290 (opens in a new tab).
  32. Chuang SK, Tian L, Wei LJ, Dodson TB. Kaplan-Meier analysis of dental implant survival: a strategy for estimating survival with clustered observations. J Dent Res. 2001;80(11):2016-20. doi:10.1177/00220345010800111301 (opens in a new tab) PMID: 11759013 (opens in a new tab).
  33. Herrmann I, Lekholm U, Holm S, Karlsson S. Impact of implant interdependency when evaluating success rates: a statistical analysis of multicenter results. Int J Prosthodont. 1999;12(2):160-6. PMID: 10371919 (opens in a new tab).
  34. Little RJA, Rubin DB. Statistical analysis with missing data. 3rd ed. Hoboken (NJ): Wiley; 2019. †
  35. Sterne JA, White IR, Carlin JB, Spratt M, Royston P, Kenward MG, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b2393. doi:10.1136/bmj.b2393 (opens in a new tab) PMID: 19564179 (opens in a new tab).
  36. Little RJ, D'Agostino R, Cohen ML, Dickersin K, Emerson SS, Farrar JT, et al. The prevention and treatment of missing data in clinical trials. N Engl J Med. 2012;367(14):1355-60. doi:10.1056/NEJMsr1203730 (opens in a new tab) PMID: 23034025 (opens in a new tab).
  37. Ratitch B, O'Kelly M, Tosiello R. Missing data in clinical trials: from clinical assumptions to statistical analysis using pattern mixture models. Pharm Stat. 2013;12(6):337-47. doi:10.1002/pst.1549 (opens in a new tab) PMID: 23292975 (opens in a new tab).
  38. Akl EA, Briel M, You JJ, Sun X, Johnston BC, Busse JW, et al. Potential impact on estimated treatment effects of information lost to follow-up in randomised controlled trials (LOST-IT): systematic review. BMJ. 2012;344:e2809. doi:10.1136/bmj.e2809 (opens in a new tab) PMID: 22611167 (opens in a new tab).
  39. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-7. doi:10.1016/S0140-6736(07)61602-X (opens in a new tab) PMID: 18064739 (opens in a new tab).
  40. Benchimol EI, Smeeth L, Guttmann A, Harron K, Moher D, Petersen I, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement. PLoS Med. 2015;12(10):e1001885. doi:10.1371/journal.pmed.1001885 (opens in a new tab) PMID: 26440803 (opens in a new tab).
  41. Klinge B, Lundström M, Rosén M, Bertl K, Klinge A, Stavropoulos A. Dental Implant Quality Register-A possible tool to further improve implant treatment and outcome. Clin Oral Implants Res. 2018;29 Suppl 18:145-151. doi:10.1111/clr.13268 (opens in a new tab) PMID: 30306699 (opens in a new tab).
  42. Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94(446):496-509. †
  43. Norman GR, Sloan JA, Wyrwich KW. Interpretation of changes in health-related quality of life: the remarkable universality of half a standard deviation. Med Care. 2003;41(5):582-92. doi:10.1097/01.MLR.0000062554.74615.4C (opens in a new tab) PMID: 12719681 (opens in a new tab).
  44. Frangakis CE, Rubin DB. Addressing an idiosyncrasy in estimating survival curves using double sampling in the presence of self-selected right censoring. Biometrics. 2001;57(2):333-42. doi:10.1111/j.0006-341x.2001.00333.x (opens in a new tab) PMID: 11414553 (opens in a new tab).
  45. Geng EH, Emenyonu N, Bwana MB, Glidden DV, Martin JN. Sampling-based approach to determining outcomes of patients lost to follow-up in antiretroviral therapy scale-up programs in Africa. JAMA. 2008;300(5):506-7. doi:10.1001/jama.300.5.506 (opens in a new tab) PMID: 18677022 (opens in a new tab).
  46. Reference verification note. 43 of 45 references are PubMed-indexed; their bibliographic records (authors, title, journal, year, volume, pages, DOI, PMID) were retrieved programmatically from the PubMed database on November 17, 2026. † Source not indexed in PubMed (book, standard, guidance or non-indexed journal); bibliographic details are cited from the published record and were not database-verified in this drafting session. Database retrieval confirms that each record exists as cited; it does not replace full-text reading by the responsible authors of every statement attributed to a source.