Skip to main content

Surface Roughness as an Exposure-Gated Risk Factor for Peri-Implantitis: A Two-Stage Hazard Hypothesis

Paper type
Hypothesis and Theory
Status
Original Work · Working Paper
Keywords:
peri-implantitis · implant surface roughness · multistate model · hazard ratio · hybrid implants · effect modification · target trial emulation
Contents
  1. 1Introduction
  2. 2Background and critical appraisal of the evidence
  3. 3Hypothesis and theoretical framework
  4. 4Proposed methodology
  5. 5Competing explanations and how the design discriminates among them
  6. 6Clinical and engineering implications
  7. 7Limitations
  8. 8Conclusions

Abstract

Background. Experimental ligature-induced peri-implantitis in dogs shows faster spontaneous bone loss at modified (moderately rough) implant surfaces than at turned surfaces, whereas clinical comparisons have not shown a stable difference in the incidence of peri-implantitis. Cohorts of surgically treated implants, in contrast, report poorer outcomes at modified surfaces. These findings are usually presented as a conflict.

Hypothesis. The conflict is reconciled if roughness has no effect on the hazard that a surface becomes exposed to the submucosal biofilm, and acts only after exposure, by raising the hazards of progressive bone loss and of relapse after treatment. A three-state formulation (surface covered, surface exposed, progressive bone loss) with transition hazards λ_(01) (independent of roughness) and λ_(12) (a function of roughness) expresses this.

Proposed approach. We derive the marginal consequences of the model, including how contrasts between surfaces depend on the prevalence of exposure, and specify four discriminating analyses: an individual-participant-data analysis of trials with a surface-by-exposure interaction, a registry target-trial emulation, a collar-design trial, and a relapse analysis in treated cohorts. An illustrative computation under stated assumptions shows that the relative contrast between surfaces stays near constant while the absolute contrast scales with exposure prevalence.

Expected contribution. The qualitative separation of initiation from progression has been reported in animal meta-analysis; the contribution here is the formal model, a corrected quantitative prediction, and explicit refutation criteria.

Conclusions. The hypothesis is testable with existing data. It is refuted if the progression hazard after exposure does not differ by surface, or if surface differences appear before exposure.

1. Introduction

Whether a moderately rough implant surface increases the risk of peri-implantitis is disputed. A systematic review that combined pre-clinical experiments with clinical studies of at least five years concluded that modified surfaces may worsen progression in experimental models, while clinical studies did not support a difference in incidence among surface types [1]. The two bodies of evidence are therefore usually read as contradictory, and clinical guidance drifts between caution about rough surfaces and indifference to the surface.

An alternative reading is that the bodies of evidence answer different questions. The animal experiments initiate disease with a ligature and then measure how far bone loss proceeds when plaque accumulates freely, so they study progression after the surface has been exposed by design. Clinical cohorts measure the first appearance of disease, which depends on whether and how early the modified surface is exposed at all. If roughness matters only for the second process, the two bodies of evidence need not disagree.

This manuscript develops that reading into a testable model. Its claims are limited. We do not propose that roughness is harmless or harmful in general. We propose a specific structure: the hazard of exposure does not depend on roughness; the hazards of progression and of relapse do. We then ask what that structure implies for marginal comparisons between surfaces, for implant designs that move the roughness boundary, and for the analyses that could refute it.

An originality search was part of the work. A systematic review of animal experiments has already reported that surfaces did not differ at initiation but differed in progression and treatment response [2]. The idea that roughness acts after the disease process has begun is therefore not new. What the literature we could reach does not contain is a formal model of the gate, a derivation of what the gate implies for incidence comparisons across cohorts with different exposure prevalence, or a set of analyses that separate the gate from competing explanations. The novelty assessment near the end of the paper returns to this point.

The evidence base was appraised at the level of PubMed abstracts only, and Embase, Scopus and Web of Science were not accessible. No patient data were analyzed. The only computations are two short scripts, whose assumptions and outputs are reported in full in Section 3.

2. Background and critical appraisal of the evidence

Experimental progression studies

The experimental line begins with ligature-induced disease in beagle dogs. In one study, implants with a sandblasted acid-etched surface and polished implants were exposed to ligatures until about 40% of supporting bone was lost; ligatures were then removed and plaque accumulation continued [3]. Later work used commercially available implants with turned, TiOblast, sandblasted acid-etched and TiUnite surfaces, again removing ligatures at about 40 to 50% bone loss and following plaque accumulation for 24 weeks [4, 5]. A subsequent two-surface experiment found that bone loss after ligature removal was significantly larger at TiUnite than at turned implants, with a longer biofilm front and a larger pocket epithelium [6].

A systematic review and meta-analysis restricted to animal models reported no statistically significant surface differences during initiation, but significant differences in progression and treatment outcomes, with turned surfaces showing less bone loss after ligature removal and more bone gain after treatment [2]. The broader review that combined animal and clinical data pooled dog experiments after the spontaneous progression phase and found greater radiographic bone loss at modified surfaces, with a pooled effect of 0.44 mm (95% CI 0.10 to 0.79; 8 publications) [1]. The same review noted that in 9 of 18 experiments no significant surface difference was seen [1].

These experiments share limitations that matter for inference. Group sizes were small (five or six dogs in the studies whose abstracts we read), ligature-induced disease differs from the slowly developing human disease, the implants within and between systems differed in macro-design as well as surface, and the endpoint (bone loss after a fixed period of unimpeded plaque accumulation) measures progression by construction. The experiments are well suited to detecting a progression effect and poorly suited to detecting an effect on exposure or onset, because exposure is imposed by the ligature.

Clinical incidence and bone-level studies

In clinical studies the picture is less consistent. The review by Stavropoulos and colleagues covered 31 publications reporting data from 7,605 patients and 26,188 implants; overall it found no significant differences in peri-implantitis incidence among surfaces, high survival (90 to 100%) up to 30 years, and no clinically relevant differences in marginal bone loss [1]. A 20-year follow-up of a randomized trial comparing a turned surface with a TiOblast surface, with 25 patients and 64 implants available, found mean bone-level changes of -0.41 mm and -0.83 mm, a difference that was not statistically significant, and the authors concluded that moderately increased roughness did not benefit long-term bone preservation [7]. In patients with severe periodontitis, a five-year randomized comparison of minimally and moderately rough implants (18 patients) found bone loss that tended to be lower at the minimally rough surface, without significant differences in pocket depth or key pathogen counts [8]. A meta-analysis of ten randomized trials favored machined surfaces for bone loss and, to a small degree, for probing depth, while rough surfaces tended toward lower failure rates [9].

Other syntheses point the opposite way on prevalence. A meta-regression of 29 studies reported prevalence of 18.5% at patient level and 12.8% at implant level, and associated moderately rough surfaces with lower prevalence (p = 0.011) [10]. A systematic review with meta-analysis of long-term crestal bone loss used bone loss as a surrogate and reported heterogeneity among surfaces [11]. A review restricted to patients with a history of periodontitis could not reach a definitive conclusion [12]. A ten-year-plus review of surface modifications found the smallest bone loss for turned surfaces but higher survival for moderately rough ones [13].

Design, sample size and bias differ widely across this set. The randomized trials are small and few implants reach the endpoint; the meta-regression is ecological, and surface type travels with implant system, era and operator; the reviews that use bone loss as a surrogate cannot separate physiological remodeling from disease. A trial designed with peri-implantitis incidence as its endpoint, comparing a hybrid implant with a fully etched implant, enrolled 112 patients with 304 implants and recorded a single case of peri-implantitis in five years, so it could not discriminate any design [14].

Treatment and recurrence

The strongest surface effects in humans appear after, not before, disease onset. In a randomized trial of adjunctive antimicrobials in surgical treatment, treatment success was 79% at implants with a nonmodified surface and 34% at implants with a modified surface, and the effect of adjunctive systemic antibiotics depended on surface [15]. In a prospective follow-up of 73 patients (130 implants) after surgical treatment, 44% of implants showed recurrence or progression, and residual deep pockets, reduced bone level and a modified surface were risk factors [16]. In a retrospective study of 149 patients, 19.9% of implants were lost during follow-up, and a modified surface predicted loss with a hazard ratio of 4.5 [17]. A 2025 systematic review of 17 clinical studies concluded that rough surfaces were generally associated with a higher risk of recurrence and implant loss than smooth surfaces, while rating certainty as low because of heterogeneity and small samples [18].

The main weakness of this set is provenance. The principal human findings come from one research line and one health-care setting, and surface is confounded with implant system and with the era in which the implants were placed [15–17]. They are best read as consistent but not independently replicated.

What the pattern of evidence implies

Three observations drive the hypothesis. First, the effect of roughness is clearest where the surface is surely exposed: after ligature-induced bone loss in dogs and after peri-implantitis has been diagnosed and treated in patients. Second, it is weakest where most implants are never exposed, namely in cohorts followed for incidence. Third, a retrospective study of 261 implants with at least five years of follow-up found that early radiographic bone loss, defined as apical migration of interproximal bone that exposed the rough surface before final prosthesis insertion, was the strongest predictor of later progressive bone loss; peri-implantitis was diagnosed at 63.2% of implants with such early loss and 5.2% of implants without it [19]. That study did not, as far as the abstract shows, stratify by surface type, so it supports the importance of exposure but not the interaction.

Peri-implantitis also behaves like a process with an early, then accelerating, course. In a Swedish sample, bone loss followed a non-linear, accelerating pattern, and most cases began within three years of function [20]. A model that treats disease as progression through ordered states with time-varying hazards is consistent with that description.

Table 1 collects the pivotal studies and our operational evidence levels.

Table 1. Pivotal evidence and operational evidence level (A to E are operational categories used in this manuscript, not GRADE: A strong, B moderate, C preliminary, D mechanistic or theoretical, E speculative hypothesis).
SourceDesign and sizeFinding relevant to the modelMain bias or limitationLevel
Albouy 2008, 2009, 2012 [4–6]Dog ligature experiments, 5 to 6 dogsMore progression at modified surfaces after ligature removalSmall n; exposure imposed; system differencesB
Garaicoa-Pazmino 2021 [2]Meta-analysis of animal studiesNo surface difference at initiation; differences in progression and treatmentAnimal-to-human extrapolationB
Stavropoulos 2021 [1]Systematic review, 7,605 patientsNo clinical incidence difference; animal progression effectHeterogeneous exposure and definitionsB
Donati 2018 [7]RCT, 20 years, 25 patientsNo bone benefit of roughnessAttrition; low powerB
Rakic 2018 [10]Meta-regression, 29 studiesLower prevalence at moderately rough surfacesEcological; confoundedC
Carcuac 2016, 2020; Romandini 2024 [15–17]RCT secondary analysis; cohortsWorse treatment outcome at modified surfacesOne research line; era and system confoundingC
Moser 2026 [19]Retrospective, 261 implantsEarly rough-surface exposure predicts progressionNot surface-stratified; unequal follow-upC
Zetterqvist 2010 [14]RCT, 304 implants, 5 yearsOne peri-implantitis case; no design contrast possibleUnderpoweredC

3. Hypothesis and theoretical framework

Statement

Exposure-gated roughness hypothesis. For an implant with a modified surface and an otherwise equivalent implant with a turned surface, placed in comparable patients and maintained in the same way, roughness has no effect on the hazard that the modified portion of the surface becomes exposed to the submucosal biofilm. After exposure, roughness raises the hazard of progressive bone loss and the hazard of recurrence after treatment. Therefore the magnitude of any marginal difference between surfaces in incidence, progression or loss depends on how many implants in the studied population reach the exposed state, and not only on the biology of the surface.

The hypothesis makes a statement about causal structure, and its strictest form is a statement about three hazards. It does not assert that every implant surface effect is mediated by exposure; it asserts that, conditional on exposure status and the covariates that determine exposure, no residual effect on the transition out of the covered state remains.

States and hazards

We define three states for an implant at time t (years since prosthesis delivery) and one treatment-related state:

  • State 0: the modified portion of the surface is covered by bone and soft tissue and not in contact with submucosal biofilm.
  • State 1: the modified portion is exposed to the submucosal biofilm, whatever the route of exposure (early remodeling, shallow placement, soft-tissue recession, or the first bone loss of an incipient lesion).
  • State 2: progressive bone loss beyond a pre-specified radiographic and clinical threshold (the clinical case definition of peri-implantitis [21]).
  • State T: treated, with disease arrested. From T the implant can return to state 1 (relapse) or stay arrested.

Let R∈{0,1} denote the surface (0 turned or machined, 1 modified), z a vector of covariates that determine exposure (placement depth, collar design, bone phenotype, bone level at loading), and u a vector of host and maintenance covariates. The transition hazards, in events per implant-year, are

λ_(01)(t∣z)=λ_(01,0)(t) exp(γ^⊤z),

λ_(12)(t∣R,u)=λ_(12,0)(t) exp(βR+δ^⊤u),

λ_(rel)(t∣R,u)=λ_(rel,0)(t) exp(β_rR+δ_r^⊤u).

The baseline hazards λ_(01,0), λ_(12,0) and λ_(rel,0) are unspecified non-negative functions of t (or of time since entry to the state, in a semi-Markov version); γ, β, δ, β_r, δ_r are log-hazard coefficients. The hypothesis is

H_0^gate:  β>0,  β_r>0,  and λ_(01) does not depend on R given z.

A relaxed alternative allows a direct path 0→2 with hazard λ_(02)(t∣R); the strict hypothesis sets λ_(02) to zero or to a value independent of R. The refutation criteria in Section 3.5 rest on that distinction.

The model has a standard formal basis in multistate event-history analysis, in which cumulative transition intensities define state-occupation probabilities [22].

Mechanism and plausibility

The proposed mechanism has three parts. First, while bone covers the modified surface, the surface is outside the oral biofilm and the host response is governed by bone remodeling and the soft-tissue seal, which depend on the design of the neck and the surgical situation. Second, once the surface is exposed, topography affects what the biofilm can do. A systematic review of clinical studies, mostly supragingival and on restorative materials, concluded that roughness above an Ra threshold of 0.2 micrometers, and higher surface free energy, facilitate biofilm formation, with direct comparisons on transmucosal implant surfaces scarce [23]. Extrapolation to submucosal surfaces is an inference. Third, the histology in the dog experiments is compatible with a larger biofilm front: at TiUnite implants the vertical dimension of the lesion and the apical extension of the biofilm were larger than at turned implants [6].

The same mechanism predicts a relapse effect. After surgical treatment the surface has by definition been exposed and contaminated, and residual biofilm on a modified surface may be harder to remove than on a turned surface. The residual-biofilm explanation for higher recurrence at modified surfaces has been proposed but, to our knowledge, not directly tested; we treat it as plausible and unproven. No evidence we reviewed measures residual biofilm on explants by surface under standard decontamination.

Marginal consequences of the gate

Marginal consequences are what make the model testable with ordinary cohort data, so we derive them. Take constant hazards: λ_(01)=a for both surfaces, λ_(12)=b_R=b_0e^(βR), no state T, and no competing losses. Time to state 2 is the sum of two independent exponential times, and for a≠b_R the probability of having reached state 2 by t is

F_R(t)=1−(b_R e^(−at)−a e^(−b_Rt))/(b_R−a).

The probability of having reached state 1 or beyond is 1−e^(−at), the same for both surfaces. Three consequences follow.

Consequence 1: the relative contrast is nearly independent of exposure prevalence, and the absolute contrast is not. Early in follow-up and for small a, F_R(t)≈ab_Rt^2/2, so the ratio F_1/F_0→b_1/b_0=e^β. For larger a the ratio drifts slightly below e^β, because exposure becomes limiting. The risk difference F_1(t)−F_0(t), by contrast, is approximately proportional to a for small a. Hence a cohort in which exposure is rare will show a small absolute difference even though the relative effect among exposed implants is unchanged. The surface contrast therefore does not vanish at low exposure; what vanishes is the absolute difference and, with it, statistical power (Section 3.6).

Consequence 2: conditional on exposure, the hazard ratio equals e^β in every cohort. Within the exposed stratum the transition hazard to state 2 is b_R regardless of how the implant became exposed, so a stratified or time-dependent analysis recovers e^β across cohorts, whereas the marginal hazard ratio of time to state 2 is not constant in time. A marginal hazard ratio compares groups whose composition changes as the more susceptible implants leave the risk set, and its interpretation as a causal effect is limited; this is a general property of hazard ratios [24, 25]. Where the effect is heterogeneous, marginal and conditional estimands can also point to different decisions [26, 27]. A study that reports only a marginal hazard ratio therefore cannot distinguish a roughness effect that is constant from one that depends on exposure.

Consequence 3: cohorts differ in the marginal contrast through a alone. Two cohorts with the same surfaces, the same b_0 and the same β but different exposure hazards a will report different risk differences and, at long horizons, somewhat different marginal hazard ratios. Heterogeneity between published cohorts in the surface contrast is then expected without any difference in surface biology. Variation in a can arise from placement depth, platform design, bone phenotype, the proportion of patients with periodontitis, or the radiographic definition of the endpoint.

Hybrid designs. In a hybrid implant the coronal zone of length c (mm) is machined and the modified surface starts below it. Let r_0 be the rate of bone loss (mm per year) after exposure at a machined surface and κ r_0 the rate at a modified surface, κ=e^β for simplicity. Bone loss L at time τ after exposure is

L(τ)=(r_0 τ,τ≤c/r_0c+κ r_0 (τ−c/r_0),τ>c/r_0)

A hybrid implant therefore behaves like a turned implant until the bone margin passes the collar, and like a modified implant afterward. In this model the collar changes where progression arrests and how fast loss proceeds beyond a given depth; it does not change the hazard that bone loss begins, which is λ_(01), and it gives no benefit in implants whose exposure is shallow or late. A collar that delays exposure of rough threads acts on λ_(01) as a design effect, not a surface effect.

Extension: dose of exposure. The strict model treats exposure as binary. A more realistic version lets λ_(12) depend on the exposed area or depth of modified surface, β=β_0+β_1A_(exp). Under this version, implants with a small area of exposure (a few thread tips) have a progression hazard close to that of a turned implant, and the surface effect grows with the extent of exposure. We keep the binary model as the primary specification because exposed area is difficult to measure radiographically.

Falsifiable predictions and refutation

P1. In a stratum of implants with radiographically confirmed exposure of the modified portion at a landmark time, the hazard ratio of progression (modified versus turned) exceeds 1 and is similar across cohorts. In implants without exposure, progression events are rare and the hazard ratio is not distinguishable from 1.

P2. The surface-by-exposure interaction on the log-hazard scale is positive: β>0 in the exposed stratum and approximately zero in the unexposed stratum.

P3. Surface differences in absolute risk of progressive bone loss are positively related to the prevalence of exposure across cohorts (meta-regression slope of risk difference on exposure prevalence greater than zero), while relative risks are roughly stable.

P4. Roughness does not alter the exposure hazard: among implants analyzed from placement, the hazard of first radiographic exposure of the modified portion is the same for turned and modified implants after adjustment for z.

P5. After treatment of peri-implantitis, the recurrence hazard is higher at modified surfaces (β_r>0), and the difference does not disappear after adjustment for implant system, placement era and residual pocket depth.

P6. Hybrid designs do not reduce the incidence of exposure of the rough surface to biofilm in cohorts followed for a limited time, but the depth of bone loss at arrest, and the rate of loss beyond the collar, are lower than for fully modified implants only if bone loss reaches the collar boundary.

P7. Within exposed implants, bone-loss rate is a graded function of surface roughness at fixed chemistry, rather than a step function.

The hypothesis is refuted, or must be substantially revised, by any of these results: (a) a progression hazard ratio near 1 in the exposed stratum with a confidence interval excluding the effect size of interest (rejecting β>0); (b) a roughness difference in λ_(01) or a non-zero roughness effect on λ_(02) after adjustment, which would imply a pre-exposure pathway; (c) a surface effect in the unexposed stratum comparable to that in the exposed stratum; (d) no association between cohort exposure prevalence and the absolute surface contrast; (e) a relapse effect that vanishes after adjustment for system and era; (f) a protective rather than harmful association of modified surfaces with progression in exposed implants, as the lower prevalence reported at moderately rough surfaces might suggest if it were causal [10].

Illustrative computation under stated assumptions

Illustrative computation under stated assumptions — not empirical data. The script two_stage_illustration.py (in the paper folder) was run in this session. Assumptions: constant hazards; b_0=0.08 per implant-year for a turned surface; hazard ratio for progression e^β=2.5 (so b_1=0.20); a 10-year horizon; four assumed values of the exposure hazard a (0.005, 0.02, 0.05 and 0.10 per implant-year), the same for both surfaces; no treatment, relapse or competing events. These values were chosen for illustration and do not come from any study. A Monte Carlo run (200,000 implants per arm, seed 28) checked the closed form at a=0.005 and a=0.05. Output is in Table 2.

Table 2. Output of the illustrative computation: cumulative probability of progressive bone loss (state 2) at 10 years by surface under the assumed parameters. Not empirical data.
Exposure hazard a (per implant-year)Exposed by 10 yTurnedModifiedRatioRisk difference (percentage points)
0.0050.0490.01530.02791.821.26
0.0200.1810.05810.10531.814.72
0.0500.3930.13150.23641.8010.49
0.1000.6320.22490.39961.7817.47

The Monte Carlo run reproduced the closed-form values (turned 0.0155 and 0.1318, modified 0.0283 and 0.2365 at a=0.005 and 0.05). It also recovered a progression rate per exposed implant-year of 0.080 (turned) and 0.202 (modified) at a=0.005, and 0.080 and 0.202 at a=0.05, a ratio of 2.54 and 2.53; the exposed-state ratio is therefore recovered in both cohorts although the cohorts differ about eightfold in their absolute differences (1.26 versus 10.49 percentage points). The ratio of cumulative incidences stays between 1.78 and 1.82, whereas the risk difference rises from 1.26 to 17.47 percentage points. A ratio of cumulative hazards, lnS_1(10)/lnS_0(10), rises from 1.83 to 2.00 over the same range.

For the hybrid geometry, under the assumed values r_0=0.10 mm per year, κ=2.5 and c=1.5 mm, computed bone loss at 5, 10, 15 and 20 years after exposure was 0.50, 1.00, 1.50 and 2.00 mm (turned), 1.25, 2.50, 3.75 and 5.00 mm (modified), and 0.50, 1.00, 1.50 and 2.75 mm (hybrid). The hybrid and turned implants coincide until the margin passes the collar, after which the hybrid implant loses bone faster. For relapse, assumed recurrence hazards of 0.05 per year (turned) and 0.10 per year (modified) give five-year recurrence probabilities of 0.221 and 0.393.

These numbers show only the internal consistency of the model; they are not estimates of any real quantity.

4. Proposed methodology

The hypothesis concerns a structure that no single existing study was designed to test. We propose four analyses of increasing causal strength and cost. All are proposals; none has been performed. Figure 2 sketches the common timeline.

Analysis A: individual-participant-data analysis of existing trials

Design. Request patient- and implant-level data from randomized trials and prospective cohorts that compared turned or minimally rough implants with modified implants for at least five years with serial standardized radiographs. Candidate sources include the trials already reported in the literature [7, 8, 14, 28–30]. A registered randomized trial of implant surface in patients with a history of periodontitis (ClinicalTrials.gov NCT05010382, completed) was found by title only and may be another source; its design and outcomes were not appraised.

Exposure variable. At prosthesis delivery and at the one-year visit, an implant is classified as exposed if the bone margin lies apical to the coronal limit of the modified surface (the first modified thread for fully modified designs, or the collar boundary for hybrids). Two calibrated readers rate a random subset of at least 10% of radiographs to estimate agreement (kappa for the binary classification, intraclass correlation for the bone-level distance).

Outcomes. Primary: first occurrence of progressive bone loss from the landmark, defined as additional bone loss beyond a pre-specified threshold together with bleeding or suppuration on probing, following the consensus case definition [21]. Secondary: implant loss; rate of bone loss in millimeters per year among exposed implants.

Analysis. Stage 1 (per trial): Cox model for time from landmark to the primary outcome with surface, exposure at landmark, and their product; implant within patient as a cluster (robust variance or shared frailty). The interaction coefficient estimates β in the exposed stratum relative to the unexposed stratum. Stage 2: random-effects pooling of trial-specific coefficients. A one-stage mixed model with trial-specific baseline hazards serves as a sensitivity analysis. Because hazard ratios are non-collapsible and depend on risk-set composition [24, 25], the report must include absolute contrasts: standardized risk differences at 5 and 10 years within exposure strata and overall.

Tests of P1 to P3. P1 and P2 are tested by the interaction and the exposed-stratum hazard ratio. P3 is tested by meta-regression of the trial-level risk difference on trial-level exposure prevalence, with the caveat that few trials will be available and the test is ecological.

Analysis B: registry target-trial emulation

Protocol of the target trial. Eligibility: adults receiving a single or multiple implants of a modified or turned surface from systems that offer both. Strategies: placement of a modified versus a turned implant. Time zero: implant placement. Outcomes: radiographic exposure of the modified portion (exposure hazard), progression after exposure, implant loss, and relapse after treatment. Causal contrast: per-protocol, with surface adherence assumed fixed. This follows the target-trial framework for observational data [31].

Emulation. In registry or health-system data with serial radiographs, emulate the protocol by weighting for baseline confounders (patient age, smoking, diabetes, periodontitis history, implant length and diameter, placement depth, bone augmentation), with implant system as a random effect, and with supportive-care adherence as a time-varying covariate. A multi-state model estimates λ_(01), λ_(12) and λ_(rel) separately by surface. P4 corresponds to the hazard ratio for the first transition; P1 and P2 to the second; P5 to the third.

Biases and their controls. Confounding by implant system and era is the main threat, because the surfaces differ with the system and with the year of placement. We propose negative-control outcomes unlikely to depend on surface after placement, such as early failure before loading, and calendar-period stratification. Misclassification of exposure is expected to attenuate the interaction; a probabilistic bias analysis based on the reader-agreement estimates would quantify this.

Analysis C: collar-design trial

A pragmatic randomized trial, or an extension of an existing trial, would randomize patients (or split-mouth sites) to hybrid versus fully modified implants of one system, to isolate the design variable. The primary outcome is not incidence of peri-implantitis, which earlier trials found too rare to detect [14, 28–30], but bone-loss depth at arrest and the proportion of implants whose bone margin crosses the collar boundary during follow-up of ten years or more. P6 predicts a difference only in implants whose bone margin crosses the boundary.

Analysis D: relapse in treated cohorts

Pool participant data from trials and cohorts of surgical or non-surgical treatment of peri-implantitis with at least three years of follow-up, stratified by surface. The primary outcome is recurrence or progression after the one-year re-evaluation, with residual probing depth at one year, implant system and baseline severity as covariates. The prediction (P5) is a recurrence hazard ratio above 1 at modified surfaces after adjustment. Prior cohorts report such effects from one research line [15–17], and a systematic review judged certainty low [18]. An independent replication in different centers is thus the key purpose of this analysis.

Optional experimental test

A factorial animal experiment would cross surface (turned or modified, same macro-design, same chemistry) with time of exposure (no ligature, ligature-induced exposure) and would measure bone loss and biofilm extent. P1 predicts a surface effect only in the exposed arms, and P7 can be tested by including surfaces that differ in roughness at fixed chemistry. Given the cost and the extrapolation problem, this is the least feasible analysis.

Measurement validity

Radiographic bone level is the shared measurement, and its limits matter. Interproximal radiographs cannot show buccal or lingual exposure; the exposure variable will therefore be misclassified in a direction that dilutes contrasts. Surface characterization should follow a common metrology protocol reporting height, spatial and hybrid parameters, because roughness and chemistry co-vary in commercial surfaces; a review has noted that most studies characterize surfaces inadequately and report only height parameters [32]. The term “turned” or “modified” is used here as a coarse proxy for roughness, and this restricts the interpretation of P7.

Sample-size reasoning

Two calculations are provided, both reproduced by sample_size.py in the paper folder; both depend on the assumed parameters of Section 3.6 and are not estimates from data.

For a marginal comparison of 10-year risk at two surfaces, the number of implants per arm for a two-sided test at α=0.05 with 80% power is

n=((z_(1−α/2)√(2p(1−p))+z_(1−β)√(p_0(1−p_0)+p_1(1−p_1)))^2)/((p_1−p_0)^2),

where p_0 and p_1 are the 10-year risks at turned and modified surfaces and p=(p_0+p_1)/2. Using the risks of Table 2, with z_(1−α/2)=1.96 and z_(1−β)=0.8416, the required numbers of implants per arm are 2,089, 528, 213 and 110 for a=0.005,0.02,0.05 and 0.10. Inflating by a design effect of 1.3 (two implants per patient and an assumed intra-patient correlation of 0.3, 1+(m−1)ρ) gives 2,715, 686, 277 and 143. A trial in a low-exposure population would therefore need roughly an order of magnitude more implants than one in a high-exposure population to detect the same relative effect, which is why such trials look null.

For the hazard ratio within the exposed stratum, the number of events needed for 1:1 allocation to detect a hazard ratio of 2.5 with the same error rates is d=4(z_(1−α/2)+z_(1−β))^2/(ln2.5)^2=37.4, rounded to 38. To test the interaction (hazard ratio 2.5 in the exposed stratum versus 1.0 in the unexposed), with equal numbers of events in each stratum, the requirement is d=8(z_(1−α/2)+z_(1−β))^2/(ln2.5)^2=74.8, rounded to 75 events per stratum. In practice the unexposed stratum will have few events, so the interaction test is limited by that stratum, and pooling across trials is needed.

Table 3. Sample-size outputs under the assumed parameters of Section 3.6 (not empirical data).
QuantityAssumed exposure hazard a = 0.0050.020.050.10
Implants per arm, marginal 10-y risk comparison2,089528213110
Same, with design effect 1.32,715686277143
Events for HR 2.5 in exposed stratum38383838
Events per stratum for interaction75757575

Ethics and feasibility

Analyses A, B and D use existing data and require data-sharing agreements and approval by the relevant ethics bodies and data custodians; no approval is claimed here. Analysis C needs a new trial protocol and registration, none of which exists. Analysis A is the most feasible and should come first.

5. Competing explanations and how the design discriminates among them

Five alternatives could produce a similar pattern of evidence. Table 4 lists the prediction that separates each from the gate hypothesis.

Table 4. Competing explanations and discriminating observations.
Competing explanationWhat it predictsDiscriminating analysisResult that favors the gate hypothesis
No roughness effect at any stage; animal result is a model artifactβ=0 in exposed implants; no relapse differenceA, B, Dβ>0 with confidence interval excluding 0
Roughness raises onset directly (pre-exposure pathway)Surface difference in λ01 or λ02; effect in unexposed stratumA, B (P2, P4)No effect in unexposed stratum; equal exposure hazards
Confounding by system, era and patientEffect disappears with system random effect and era stratificationB, DEffect persists within system
Exposure is a marker of host susceptibility, not a gateExposure predicts progression equally at both surfacesA (interaction)Interaction positive; main effect of exposure alone is insufficient
Chemistry or nanotopography, not roughness, carries the effectEffect tracks coating or treatment, not Sa or SdrOptional animal test; metrologyEffect graded by roughness at fixed chemistry

The fourth alternative is the most difficult to exclude. Early bone loss that exposes the rough surface predicts later progression and peri-implantitis [19], but it is also a plausible marker of a susceptible host or a poorly adapted soft-tissue seal. The gate hypothesis requires a surface-by-exposure interaction, not an exposure main effect: if exposure were only a marker of susceptibility, turned implants would show the same exposure-associated risk as modified ones. The effect of exposure is thus compared across surfaces, not within one surface.

The second alternative has a specific design feature. Clinical studies that measure bone loss in the first year or before loading provide a window in which most implants are unexposed; if surface differences in early bone loss appear there, the exposure hazard depends on roughness. Randomized trials of rough versus turned surfaces that report early remodeling would contribute to this test. A related long-term review of surface roughness and patient factors on peri-implant bone loss concluded that the clinical impact of roughness alone on bone loss and peri-implantitis risk seems limited, with patient factors such as periodontitis history and smoking mattering more [33]; this is compatible with either the gate or the null.

The third alternative can be partly addressed in the registry emulation if several systems offer both surfaces, because system then becomes a stratification variable.

6. Clinical and engineering implications

These implications are conditional on the hypothesis surviving the tests in Section 4 and should not guide practice now.

Surface choice would be a question about exposure risk. If the gate holds, the relevant clinical question is not whether a modified surface raises peri-implantitis risk in general, but how likely the modified portion is to be exposed in a given patient and site. In patients and sites at low exposure risk (adequate bone, apical placement relative to the surface boundary, thick mucosa, good maintenance), the absolute cost of a modified surface would be small. Where exposure is likely, as in patients with periodontitis history, thin phenotype or limited maintenance, the absolute cost would be larger, and a smoother coronal zone or a turned implant would be more attractive. The randomized evidence in periodontitis-history patients is small and did not settle this [8, 12].

Hybrid designs would be judged on arrest depth, not incidence. The model predicts that a machined collar shifts the point where bone loss arrests when the margin crosses the collar, and has no influence on whether bone loss starts. The short randomized comparisons of hybrid designs showed no differences in clinical outcomes [28–30], which agrees with a low event rate rather than with a failure of the design. Reporting the proportion of implants that cross the collar boundary and the depth at arrest would make such trials informative.

Treatment protocols could be stratified by surface. If β_r>0, modified surfaces would need more intensive decontamination, closer supportive care after therapy, or an earlier decision to remove implants, and trials of decontamination methods should stratify by surface. Existing reviews suggest that the effect of adjunctive measures may depend on the surface [15, 18], but this rests on a single trial line and low-certainty evidence.

Engineering. The design variable of interest is the position of the roughness boundary relative to the expected bone margin, together with the transition between machined and modified zones. A model of this kind gives a design target: the expected distribution of bone-margin position over time under a given placement protocol and the fraction of implants whose margin reaches the boundary. It does not specify an optimal roughness.

Reinterpreting the existing literature. Between-study variation in the surface contrast would be expected from differences in exposure prevalence alone. Observed differences between brands in the odds of moderate or severe peri-implantitis in a Swedish population study [34] and the low patient-level frequency of peri-implantitis reported in a meta-analysis of prospective studies of a single moderately rough surface [35] would then be partly attributable to design and placement features that govern exposure, as well as to surface. This is a possibility, not a finding.

7. Limitations

  • Evidence appraisal. All evidence was read at abstract level, and Embase, Scopus and Web of Science were not searched. Study characteristics, outcome definitions and adjustment sets could not be verified.
  • No empirical test. The model has not been fitted to data. The computed numbers are consequences of assumed parameters.
  • Constant-hazard simplification. Real hazards change over time, and peri-implantitis accelerates after onset [20]. With time-varying hazards the ratio of cumulative incidences would not stay near constant; the model permits this, but the closed forms do not.
  • Exposure is difficult to observe. Bone margin relative to the surface boundary is visible on interproximal radiographs only, and the first exposure may be buccal. Misclassification would bias interaction estimates toward the null.
  • Exposure may be endogenous. The same host susceptibility that leads to early bone loss may increase progression. Only the surface-by-exposure interaction, and not the main effect of exposure, is specific to the hypothesis.
  • Roughness is a proxy. Turned and modified implants differ in chemistry, nanotopography, residual contamination and often macro-design. The model attributes the progression effect to “modified surface,” which may not be roughness as such.
  • Binary exposure. The strict model does not describe partial exposure, and the extension with exposed area is not specified further.
  • Competing risks and treatment. Implant removal, prosthetic complications, and early biological complications compete with progression; they are not modeled. Treatment in the model leads only to arrest or relapse.
  • Prior contradictory evidence. A meta-regression associated moderately rough surfaces with lower prevalence [10], and several syntheses found higher survival for rough surfaces [9, 13]. The model can accommodate lower early failure through osseointegration benefits, but this is an additional assumption and not an implication.
  • Provenance of treatment data. The human relapse evidence comes mostly from one research line and has not been independently replicated [15–17].
  • Animal-to-human extrapolation. The dog model forces exposure and tests progression, so it cannot distinguish the gate from a general progression effect.

8. Conclusions

The conflict between animal and clinical evidence on moderately rough surfaces may be a conflict between questions: progression after exposure and incidence in populations where exposure is uncommon. We propose that roughness raises the hazards of progression and relapse only after the surface is exposed. A closed-form calculation under assumed parameters shows that, if this holds, the relative contrast between surfaces would remain nearly constant across cohorts while the absolute contrast and the statistical power would scale with exposure prevalence; null findings from low-exposure cohorts would then be uninformative and not reassuring. The hypothesis is refuted if progression after documented exposure does not differ by surface, or if surface effects appear before exposure. A surface-by-exposure interaction analysis of existing trial data is the most feasible first test.

Cite this paper

Jain S, Gupta A. Surface Roughness as an Exposure-Gated Risk Factor for Peri-Implantitis: A Two-Stage Hazard Hypothesis. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/surface-roughness-exposure-gated-risk

References

  1. Stavropoulos A, Bertl K, Winning L, Polyzois I. What is the influence of implant surface characteristics and/or implant material on the incidence and progression of peri-implantitis? A systematic literature review. Clin Oral Implants Res. 2021;32 Suppl 21:203-229. doi:10.1111/clr.13859 (opens in a new tab). PMID: 34642989 (opens in a new tab).
  2. Garaicoa-Pazmino C, Lin GH, Alkandery A, Parra-Carrasquer C, Suárez-López Del Amo F. Influence of implant surface characteristics on the initiation, progression and treatment outcomes of peri-implantitis: A systematic review and meta-analysis based on animal model studies. Int J Oral Implantol (Berl). 2021;14(4):367-382. PMID: 34726847 (opens in a new tab).
  3. Berglundh T, Gotfredsen K, Zitzmann NU, Lang NP, Lindhe J. Spontaneous progression of ligature induced peri-implantitis at implants with different surface roughness: an experimental study in dogs. Clin Oral Implants Res. 2007;18(5):655-61. doi:10.1111/j.1600-0501.2007.01397.x (opens in a new tab). PMID: 17608738 (opens in a new tab).
  4. Albouy JP, Abrahamsson I, Persson LG, Berglundh T. Spontaneous progression of peri-implantitis at different types of implants. An experimental study in dogs. I: clinical and radiographic observations. Clin Oral Implants Res. 2008;19(10):997-1002. doi:10.1111/j.1600-0501.2008.01589.x (opens in a new tab). PMID: 18828815 (opens in a new tab).
  5. Albouy JP, Abrahamsson I, Persson LG, Berglundh T. Spontaneous progression of ligatured induced peri-implantitis at implants with different surface characteristics. An experimental study in dogs II: histological observations. Clin Oral Implants Res. 2009;20(4):366-71. doi:10.1111/j.1600-0501.2008.01645.x (opens in a new tab). PMID: 19309770 (opens in a new tab).
  6. Albouy JP, Abrahamsson I, Berglundh T. Spontaneous progression of experimental peri-implantitis at implants with different surface characteristics: an experimental study in dogs. J Clin Periodontol. 2012;39(2):182-7. doi:10.1111/j.1600-051X.2011.01820.x (opens in a new tab). PMID: 22136592 (opens in a new tab).
  7. Donati M, Ekestubbe A, Lindhe J, Wennström JL. Marginal bone loss at implants with different surface characteristics - A 20-year follow-up of a randomized controlled clinical trial. Clin Oral Implants Res. 2018;29(5):480-487. doi:10.1111/clr.13145 (opens in a new tab). PMID: 29569767 (opens in a new tab).
  8. Raes M, D’hondt R, Teughels W, Coucke W, Quirynen M. A 5-year randomized clinical trial comparing minimally with moderately rough implants in patients with severe periodontitis. J Clin Periodontol. 2018;45(6):711-720. doi:10.1111/jcpe.12901 (opens in a new tab). PMID: 29611221 (opens in a new tab).
  9. Hussein A, Shah M, Atieh MA, Alhimairi S, Amir-Rad F, Elbishari H. Influence of Implant Surfaces on Peri-Implant Diseases - A Systematic Review and Meta-Analysis. Int Dent J. 2025;75(1):75-85. doi:10.1016/j.identj.2024.10.007 (opens in a new tab). PMID: 39532569 (opens in a new tab).
  10. Rakic M, Galindo-Moreno P, Monje A, Radovanovic S, Wang HL, Cochran D, et al. How frequent does peri-implantitis occur? A systematic review and meta-analysis. Clin Oral Investig. 2018;22(4):1805-1816. doi:10.1007/s00784-017-2276-y (opens in a new tab). PMID: 29218422 (opens in a new tab).
  11. Doornewaard R, Christiaens V, De Bruyn H, Jacobsson M, Cosyn J, Vervaeke S, et al. Long-Term Effect of Surface Roughness and Patients’ Factors on Crestal Bone Loss at Dental Implants. A Systematic Review and Meta-Analysis. Clin Implant Dent Relat Res. 2017;19(2):372-399. doi:10.1111/cid.12457 (opens in a new tab). PMID: 27860171 (opens in a new tab).
  12. Dank A, Aartman IHA, Wismeijer D, Tahmaseb A. Effect of dental implant surface roughness in patients with a history of periodontal disease: a systematic review and meta-analysis. Int J Implant Dent. 2019;5(1):12. doi:10.1186/s40729-019-0156-8 (opens in a new tab). PMID: 30756245 (opens in a new tab).
  13. Wennerberg A, Albrektsson T, Chrcanovic B. Long-term clinical outcome of implants with different surface modifications. Eur J Oral Implantol. 2018;11 Suppl 1:S123-S136. PMID: 30109304 (opens in a new tab).
  14. Zetterqvist L, Feldman S, Rotter B, Vincenzi G, Wennström JL, Chierico A, et al. A prospective, multicenter, randomized-controlled 5-year study of hybrid and fully etched implants for the incidence of peri-implantitis. J Periodontol. 2010;81(4):493-501. doi:10.1902/jop.2009.090492 (opens in a new tab). PMID: 20367092 (opens in a new tab).
  15. Carcuac O, Derks J, Charalampakis G, Abrahamsson I, Wennström J, Berglundh T. Adjunctive Systemic and Local Antimicrobial Therapy in the Surgical Treatment of Peri-implantitis: A Randomized Controlled Clinical Trial. J Dent Res. 2016;95(1):50-7. doi:10.1177/0022034515601961 (opens in a new tab). PMID: 26285807 (opens in a new tab).
  16. Carcuac O, Derks J, Abrahamsson I, Wennström JL, Berglundh T. Risk for recurrence of disease following surgical therapy of peri-implantitis-A prospective longitudinal study. Clin Oral Implants Res. 2020;31(11):1072-1077. doi:10.1111/clr.13653 (opens in a new tab). PMID: 32870513 (opens in a new tab).
  17. Romandini M, Bougas K, Alibegovic L, Hosseini S, Carcuac O, Berglundh T, et al. Long-term outcomes and prognostic factors of surgical treatment of peri-implantitis - A retrospective study. Clin Oral Implants Res. 2024;35(3):321-329. doi:10.1111/clr.14228 (opens in a new tab). PMID: 38112108 (opens in a new tab).
  18. Gardelis P, Giannopoulou C, Stavropoulos A, Zekeridou A. Impact of implant surface modifications on long-term outcome of surgical peri-implantitis treatment: a systematic review. Front Dent Med. 2025;6:1661369. doi:10.3389/fdmed.2025.1661369 (opens in a new tab). PMID: 41069513 (opens in a new tab).
  19. Moser D, Ozgur E, Tjokro J, Raabe C, Chappuis V, Avila-Ortiz G, et al. Effect of Early Radiographic Bone Loss Before Loading on Implant-Related Outcomes: A Long-Term Retrospective Study. Clin Oral Implants Res. 2026;37(6):704-713. doi:10.1111/clr.70112 (opens in a new tab). PMID: 41742397 (opens in a new tab).
  20. Derks J, Schaller D, Håkansson J, Wennström JL, Tomasi C, Berglundh T. Peri-implantitis - onset and pattern of progression. J Clin Periodontol. 2016;43(4):383-8. doi:10.1111/jcpe.12535 (opens in a new tab). PMID: 26900869 (opens in a new tab).
  21. 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).
  22. Andersen PK, Keiding N. Multi-state models for event history analysis. Stat Methods Med Res. 2002;11(2):91-115. doi:10.1191/0962280202SM276ra (opens in a new tab). PMID: 12040698 (opens in a new tab).
  23. Teughels W, Van Assche N, Sliepen I, Quirynen M. Effect of material characteristics and/or surface topography on biofilm development. Clin Oral Implants Res. 2006;17 Suppl 2:68-81. doi:10.1111/j.1600-0501.2006.01353.x (opens in a new tab). PMID: 16968383 (opens in a new tab).
  24. Abrahamowicz M, Beauchamp ME, Roberts EK, Taylor JMG. Revisiting the hazards of hazard ratios through simulations and case studies. Eur J Epidemiol. 2025;40(6):611-629. doi:10.1007/s10654-025-01245-6 (opens in a new tab). PMID: 40608267 (opens in a new tab).
  25. Post RAJ, van den Heuvel ER, Putter H. The built-in selection bias of hazard ratios formalized using structural causal models. Lifetime Data Anal. 2024;30(2):404-438. doi:10.1007/s10985-024-09617-y (opens in a new tab). PMID: 38358572 (opens in a new tab).
  26. Phillippo DM, Remiro-Azócar A, Heath A, Baio G, Dias S, Ades AE, et al. Effect modification and non-collapsibility together may lead to conflicting treatment decisions: A review of marginal and conditional estimands and recommendations for decision-making. Res Synth Methods. 2025;16(2):323-349. doi:10.1017/rsm.2025.2 (opens in a new tab). PMID: 41626969 (opens in a new tab).
  27. Whitcomb BW, Naimi AI. Defining, Quantifying, and Interpreting “Noncollapsibility” in Epidemiologic Studies of Measures of “Effect”. Am J Epidemiol. 2021;190(5):697-700. doi:10.1093/aje/kwaa267 (opens in a new tab). PMID: 33305812 (opens in a new tab).
  28. Serrano B, Sanz-Sánchez I, Serrano K, Montero E, Sanz M. One-year outcomes of dental implants with a hybrid surface macro-design placed in patients with history of periodontitis: A randomized clinical trial. J Clin Periodontol. 2022;49(2):90-100. doi:10.1111/jcpe.13575 (opens in a new tab). PMID: 34881442 (opens in a new tab).
  29. Serrano B, Sanz-Sánchez I, Montero E, Sanz M. Three-Year Outcomes of Dental Implants With a Hybrid Surface Macro-Design Placed in Patients With History of Periodontitis: A Randomised Clinical Trial. J Clin Periodontol. 2025;52(6):802-812. doi:10.1111/jcpe.14153 (opens in a new tab). PMID: 40277088 (opens in a new tab).
  30. Glibert M, Vervaeke S, Ibrahim W, Doornewaard R, De Bruyn H. A Split-Mouth Study to Assess the Effect of Implant Surface Roughness on Implant Treatment Outcome After 5 Years. Int J Periodontics Restorative Dent. 2023;43(1):113-119. doi:10.11607/prd.6264 (opens in a new tab). PMID: 36661886 (opens in a new tab).
  31. Hernán MA, Wang W, Leaf DE. Target Trial Emulation: A Framework for Causal Inference From Observational Data. JAMA. 2022;328(24):2446-2447. doi:10.1001/jama.2022.21383 (opens in a new tab). PMID: 36508210 (opens in a new tab).
  32. Wennerberg A, Albrektsson T. Effects of titanium surface topography on bone integration: a systematic review. Clin Oral Implants Res. 2009;20 Suppl 4:172-84. doi:10.1111/j.1600-0501.2009.01775.x (opens in a new tab). PMID: 19663964 (opens in a new tab).
  33. De Bruyn H, Christiaens V, Doornewaard R, Jacobsson M, Cosyn J, Jacquet W, et al. Implant surface roughness and patient factors on long-term peri-implant bone loss. Periodontol 2000. 2017;73(1):218-227. doi:10.1111/prd.12177 (opens in a new tab). PMID: 28000269 (opens in a new tab).
  34. Derks J, Schaller D, Håkansson J, Wennström JL, Tomasi C, Berglundh T. Effectiveness of Implant Therapy Analyzed in a Swedish Population: Prevalence of Peri-implantitis. J Dent Res. 2016;95(1):43-9. doi:10.1177/0022034515608832 (opens in a new tab). PMID: 26701919 (opens in a new tab).
  35. Karl M, Albrektsson T. Clinical Performance of Dental Implants with a Moderately Rough (TiUnite) Surface: A Meta-Analysis of Prospective Clinical Studies. Int J Oral Maxillofac Implants. 2017;32(4):717-734. doi:10.11607/jomi.5699 (opens in a new tab). PMID: 28708905 (opens in a new tab).