A Load-Quantified Threshold Hypothesis for Loss of Osseointegration Under Occlusal Overload
- Paper type
- Hypothesis and Theory
- Status
- Original Work · Working Paper
- Keywords:
- occlusal overload · osseointegration · bending moment · mechanostat · animal models · dose-response
Contents
Abstract
Background. Animal experiments on occlusal overload of osseointegrated implants disagree. One primate model reported loss of osseointegration under lateral supra-occlusal load, whereas several canine models, with plaque controlled, reported no loss of integration or marginal bone. Load in these studies was usually defined by occlusal height or contact, not by force, and the strain value of about 3000 microstrain used as a “pathological overload” limit comes from long-bone mechanostat theory.
Hypothesis. In the absence of peri-implant inflammation, loss of osseointegration occurs only when the crestal bending moment (or the peak crestal bone strain it produces) exceeds a threshold M* and the number of loaded cycles exceeds a threshold that decreases as the moment rises. The conflicting models lie on opposite sides of M*. The 3000-microstrain figure is not validated for peri-implant bone.
Proposed approach. We state the exposure explicitly as a bending moment at the bone crest, M = F·h, with a dimensional strain index, and compute illustrative moments for stated force and lever-arm ranges. We list, for each animal study, which inputs are known and which are unknown. We propose a large-animal experiment with instrumented abutments, randomized moment levels, fixed cycle exposure, controlled plaque, and a pre-specified dose–response and change-point analysis, and we supply a Monte Carlo power illustration. Eight numbered predictions and refuting results are given.
Expected contribution. The manuscript converts a qualitative dispute into a test with a common dose axis, and it narrows the claim of prior reviews that load was unquantified: a few studies did quantify it, but none placed the models on one scale.
Conclusions. The hypothesis is untested and may be false. A graded response without a threshold, an effect driven by load direction, or occult inflammation would each refute it. Appraisal was limited to abstracts.
1. Introduction
Whether occlusal overload can by itself destroy the bone-implant interface has been argued since the first decade of osseointegration. Clinical opinion, textbooks, and prosthetic design rules assume that it can. The controlled animal evidence that is supposed to settle the question points in two directions. Isidor reported that five of eight laterally overloaded implants in four monkeys lost osseointegration, with mobility and radiolucency appearing between 4.5 and 15.5 months after overload began, whereas implants exposed only to plaque lost bone but remained integrated [1]. A dog study with six animals and 45 evaluable implants found no difference in marginal bone level or bone-to-implant contact after eight months of supra-occlusal crowns with plaque control [2]. A second canine study found higher bone-to-implant contact and only slight crestal resorption under overload in uninflamed mucosa, but aggravated bone loss when plaque-induced inflammation was present [3]. A third dog study, using cantilevered reconstructions, found no loss of osseointegration, a gain in radiographic bone density on one surface, and more technical complications under excessive load [4].
Two systematic reviews and a critical review have summarized this record. The review by Naert and colleagues posed a question on overload and bone or implant loss and could not answer it from the available literature [5]. Duyck and Vandamme later identified the lack of quantification of so-called overload at the implant level as one of the main shortcomings, and judged the evidence weak and not supportive of overload causing bone loss except with inflammation [6]. Their diagnosis is the starting point of the present manuscript. It is also the reason this manuscript does not claim that load quantification is a new idea. The claim is narrower. If load is the missing variable, the published models can be re-expressed on a single mechanical scale, with the unknown inputs declared, and a threshold model can be written so that it makes risky predictions. That has not, to our knowledge, been done.
A second problem sits beside the first. According to our source notes, simulations and occlusal teaching often adopt the mechanostat of Frost, in which strain above roughly 3000 microstrain marks a pathological overload window [7]. That value was framed for load-bearing skeletal bone. A direct test in a canine overload model measured bite load and then bone strain and reported strains far below it [8]. If the figure does not apply to peri-implant bone, the field has no validated numerical threshold at all.
This paper proposes the following chain. Occlusal force acting at a lever arm produces a bending moment at the crest. The moment, through implant geometry and bone stiffness, produces a crestal strain. The strain, repeated over a sufficient number of cycles and absent inflammation, either remains in an adaptive range or crosses into damage. The hypothesis is that the disagreement among animal models is a disagreement about where each experiment sat on this scale. The evidence base is abstract-level only, no experiment was run, and the arithmetic provided is illustrative.
2. Background and critical appraisal of the evidence
Origin of the strain thresholds
The mechanostat hypothesis describes how load-bearing bones in postnatal vertebrates adapt their strength to the mechanical loads placed on them, through feedback on tissue-level strain [7]. A companion clinical-oriented update restated the same framework and its relation to Wolff’s law for clinicians [9]. The strain windows commonly quoted in implant papers (disuse, adapted, mild overload, pathological overload) are attributed to this framework. The abstract of the 2003 update, as retrieved, does not give a peri-implant strain value; we therefore treat the application of a fixed pathological window to alveolar bone around implants as an extrapolation made in later work, not as a finding of the original theory. Stanford and Brand discussed implant occlusion in terms of strain-adaptive remodeling [10]. A recent narrative review offers a hypothesis-generating framework for osteocyte mechanobiology in peri-implant bone adaptation, which indicates continuing interest in a mechanistic account but not a validated threshold [11].
Remodeling simulations
Several numerical models embed an overload term: one remodeling model added a resorption response and was demonstrated by simulation [12], and others predicted remodeling around implant systems [13, 14]. Comparisons of numerical models and of candidate mechanical stimuli addressed validation and parameter sensitivity [15, 16]. The abstracts of the simulation papers we retrieved report predictions and sensitivity analyses, not a calibrated overload threshold validated against failure of osseointegration. Simulation papers therefore cannot be used as independent evidence for a numerical threshold; their thresholds are inputs.
Animal overload studies
The controlled animal evidence is small. Table 1 lists the studies that matter most with their design, size, quantification of load, and main bias. The following paragraphs discuss them in turn.
Monkeys. Isidor’s study placed five screw-type titanium implants in each of four Macaca fascicularis, restored one lateral segment with a prosthesis in supra-occlusal contact against a maxillary splint, and replaced the prosthesis during the experiment [1]. The renewed prosthesis displaced the mandible laterally, so the load became lateral rather than axial. The abstract does not report the magnitude of force, the lever arm, or the displacement. Five of eight overloaded implants lost osseointegration [1]. The control implants were maintained differently (a plaque-accumulation regimen on separate implants), so contrasts between overload and plaque are confounded with implant position and with the hygiene regimen.
A separate monkey series by Miyata and colleagues used superstructures defined by excess occlusal height. In the part reviewed here, the heights were 100, 180, and 250 micrometers, and bone resorption tended to increase from 180 micrometers or more of excess height while inflammation was kept absent [17]. The abstract describes an inflammation-free state but gives no force values. A height is a displacement, not a load: the force produced by a given interference depends on the stiffness of the antagonist, the periodontal ligament of the opposing tooth, and the mandibular motion. The threshold of 180 micrometers is therefore a statement about a particular preparation, and we found no replication in the abstracts we read.
Dogs. The dog studies have good plaque control, tens of implants, and histology, but none of the abstracts reports applied force. Heitz-Mayfield and colleagues found marginal bone levels of 3.6 mm in controls and 3.7 mm in loaded implants and bone-to-implant contact of 73% and 74% [2]. Kozlovsky and colleagues, with four beagles, found that loading increased contact in uninflamed sites and aggravated loss in inflamed sites [3]. Lima and colleagues used cantilever units and reported more abutment fractures (23.3% of units overall) and loosened screws, which indicates that the mechanical system was stressed even though bone was not damaged [4].
Two canine studies applied static lateral load. Gotfredsen and colleagues, using an expansion-screw appliance after experimental mucositis or peri-implantitis, studied lateral static load in dogs [18]. Podaropoulos and colleagues applied progressive static forces of 100, 200, and 300 g through springs in three dogs with 24 implants and examined bone histologically [19]. Our source notes summarize these static loads as harmless or anabolic; the abstract text we retrieved did not include the results, so that summary needs checking against the full texts. Static load is not a stand-in for cyclic occlusal load, and we do not treat these as tests of the hypothesis at high cycle numbers.
Rodent and rabbit. Duyck and colleagues provide the only animal study identified that states the moment explicitly. In rabbit tibiae, a transverse force of 29.4 N at 1.5 mm from the implant top gave a bending moment of 4.4 N·cm statically, and a force of 14.7 N at 50 mm gave 73.5 N·cm, applied for a total of 2,520 cycles at 1 Hz over 14 days; our source notes record crater-like marginal bone reactions only for the dynamic condition [20]. This is the nearest published dose-quantified evidence, and it also contains the cycle dimension. Its limitations are that the tibia is not alveolar bone, the loading was imposed by an actuator, the duration was short, and the relation between rabbit implant moments and human occlusal moments is unknown. A rat model reported degenerative bone change around an overloaded implant without infection [21].
What was measured: forces and strains
Load measurement around implants is technically possible. A review of devices mounted on implants for measuring three-dimensional functional load in vivo documents that the necessary instrumentation exists, although the abstract reports that information on in vivo load is limited [22]. A human in vivo study measured the magnitude and distribution of occlusal forces on implants supporting fixed prostheses [23]. In dogs and cats, direct in vivo bite measurements have been highly variable with animal volition and technique, and indirect methods (electrical stimulation, calculation, finite element models) do not necessarily represent natural bite [24]. In macaques, bone strain has been used as an indicator of bite force [25], and strain during mastication has been combined with finite element modeling of the mandible [26]. A pilot study described an in vivo strain-gauge technique for implant-to-bone force transmission [27]. These precedents mean that the proposed instrumentation is feasible in principle. They do not show that anyone has recorded abutment moment and crestal bone outcome together in an overload model of the type discussed here.
The only study that quantified occlusal load in an overload model and then measured strain is Kan and colleagues [8]. Two greyhounds received non-splinted crowns in supra-occlusal contact for eight weeks; a bite-force device recorded mean and peak occlusal loads of 434 and 795 N; bonded rosette gauges, applied in vitro after sacrifice with loads up to 476 N, recorded up to 1133 microstrain at implant apices and 229 microstrain at distant sites; and linear extrapolation placed the 3000-microstrain level at 1344 N, above the highest measured in vivo load [8]. The study has two animals and eight weeks of loading, measured strain in vitro on explanted bone, and extrapolated beyond the loading range. The abstract does not report whether bone was lost, because the question was strain. It is a useful anchor, not a refutation of a peri-implant threshold, because it concerns whether one particular threshold was reached in one preparation. We return to its use for calibration in Section 3.
| Study | Model and size | How load was defined | Force or moment reported in abstract? | Main bias or limitation | Level |
|---|---|---|---|---|---|
| Isidor 1996 [1] | 4 monkeys; 8 overloaded implants; 5 lost integration | Supra-occlusal prosthesis; lateral displacement after prosthesis renewal | No | Tiny n; clustered implants; different hygiene regimen for comparison implants | C |
| Miyata 2000 [17] | Monkeys; excess height 100, 180, 250 micrometers | Excess occlusal height | No | Height is not force; single group; no replication found | C |
| Heitz-Mayfield 2004 [2] | 6 dogs; 45 implants; 8 months | Supra-occlusal gold crowns | No | Unquantified load; one short follow-up; healthy mucosa by design | C |
| Kozlovsky 2007 [3] | 4 beagles; 12 months; inflamed and uninflamed arms | Supra-occlusal contact versus infra-occlusal | No | Very small n; machined implants only | C |
| Lima 2019 [4] | 5 dogs; cantilever versus single crown | Excessive occlusal contacts on cantilever | No | Weak mechanical definition; technical failures may have limited bone load | C |
| Duyck 2001 [20] | 10 rabbits; tibia; 14 days | Actuator force and lever arm | Yes (4.4 and 73.5 N·cm; 2,520 cycles) | Non-alveolar bone; short; actuator loading | C |
| Kan 2014 [8] | 2 greyhounds; 8 weeks | Bite-force device; strain in vitro | Yes (mean 434 N, peak 795 N) | n = 2; in vitro strain; extrapolation to 1344 N | C |
| Gotfredsen 2002 [18]; Podaropoulos 2016 [19] | Dogs; static lateral or orthodontic load | Spring or screw force | Partly (100–300 g in [19]) | Static, not cyclic; inflamed arm in [18] | C |
| Mojaver 2025 [28]; Di Fiore 2022 [29] | Systematic reviews; 80 and 7 studies | Mixed designs; occlusal analysis methods | Variable | Heterogeneous exposure; moderate-to-serious bias in the included studies of [29] | B |
Note. Evidence levels are operational categories used in this manuscript (A strong, B moderate, C preliminary, D mechanistic or theoretical, E speculative hypothesis), not GRADE ratings.
Clinical evidence and the exposure problem
Clinical studies cannot place implants on a force scale, because clinical exposure is usually a proxy. A systematic review of overload and peri-implant health pooled 80 studies of mixed design [28]. A review restricted to occlusal analysis through digital and analogic methods found only seven studies, four with a positive association, one using a digital device, and all with moderate-to-serious bias [29]. A systematic review of traumatic occlusal forces and peri-implant bone loss reached a more cautious view [30]. A cross-sectional study of 916 implants reported higher odds of peri-implantitis with wear facets (an odds ratio of about 2.4, according to our source notes) [31], while an earlier study found no effect of occlusal wear on bone loss rate [32]. In a 15-year prospective study of mandibular fixed prostheses, bite force, clenching, and cantilever length were minor determinants compared with smoking and hygiene [33]. The peri-implantitis workshop review, according to our source notes, states that progressive bone loss without inflammation is rare [34]. One commentary read the evidence as showing no association between overload and tissue loss under healthy conditions [35].
This literature does not contradict the hypothesis: uncalibrated proxies cannot show where clinical exposure sits relative to any threshold, and an association can arise because load aggravates inflammatory bone loss [3, 6].
Why the dispute persists
Five features of the record are enough to explain a persistent disagreement without invoking error by any group.
- Dose is undefined. Occlusal height, contact, or cantilever length is not force, and the same height may produce very different bending moments depending on antagonist stiffness and mandibular motion.
- Direction differs. Isidor’s model became lateral after prosthesis renewal [1]; whether the dog models imposed substantial non-axial load cannot be read from their abstracts.
- Cycles differ. Experiments last from days to more than a year, and chewing frequency differs among species.
- Inflammation differs. Studies that kept plaque under control and studies that did not are not comparable; the sole direct comparison found aggravation of loss only with inflammation [3].
- Species and bone differ. Primate, canine, and lagomorph bone differ in remodeling rate, mandibular loading, and tooth geometry. Dog bite force is highly variable between animals and between measurement methods [24].
The hypothesis addresses the first three directly and treats the fourth and fifth as design controls.
3. Hypothesis and theoretical framework
Formal statement
Let M be the peak bending moment about the bone crest produced by a single loading cycle of an implant-supported restoration, N the number of loaded cycles at that moment, and I a binary indicator of peri-implant inflammation (clinical or histological). The hypothesis (H1) states that:
- When I = 0, loss of osseointegration (defined as clinical mobility, a continuous radiolucency, or histologic loss of bone-to-implant contact beyond a pre-specified fraction) occurs with negligible probability when M < M*(N), and with a probability that rises steeply once M exceeds M*(N).
- M*(N) is a non-increasing function of N: a larger moment is tolerated for few cycles and a smaller moment is tolerated for many.
- When I = 1, the threshold is lower: M*_inflamed(N) < M*(N).
- The apparent conflict between the Isidor monkeys and the Heitz-Mayfield, Kozlovsky, and Lima dogs arises because their M, N, or both lay on opposite sides of M*(N).
- A single strain value, such as 3000 microstrain, does not define M* across implant geometries and species.
A competing graded form, in which risk rises smoothly without a transition, is retained as the principal alternative (Section 5).
Assumptions
- A1. Inflammation can be controlled or measured well enough to classify sites. This is the weakest assumption in the animal literature, because peri-implant mucosa may be subclinically inflamed.
- A2. Within one species, force at the abutment can be recorded or reconstructed to an acceptable error. Section 4 specifies instrumentation.
- A3. The relevant mechanical exposure at the crest is dominated by bending and not by axial load. This is plausible for lateral and cantilever loading, and may be wrong for axial overload.
- A4. Bone stiffness at the crest and the implant diameter are known or can be measured.
Mathematics
Moment. For a lateral force component F_lat (newtons) acting at height h (millimeters) above the bone crest, the moment about the crest is
M=F_(lat)⋅h
With a total force F inclined at angle θ to the implant axis, and an axial component acting at offset a from the axis, the moment is M=F(hsinθ+acosθ). Here F_lat = F sin θ. Values are reported in N·cm, with 1 N·cm = 10 N·mm. This is statics and requires no assumption about bone.
Strain index. For a solid cylinder of diameter D (mm) subjected to a moment M (N·mm), the nominal surface bending stress is σ = M/Z, with section modulus Z = πD³/32. Dividing by an elastic modulus E (N/mm², equal to MPa) gives a nominal strain. We introduce a dimensionless constant κ to absorb everything that the cylinder idealization ignores (embedment in bone, load sharing with the cortical plate, soft tissue seal, stress concentration):
ε^*=(κ M)/(E D^3), κ_0=32/π
The value κ₀ = 32/π corresponds to the surface stress of an unsupported solid cylinder. It is a scaling bound, not a prediction of crestal bone strain, because surrounding bone shares the load and a cylinder in bone is not a free cantilever. We therefore treat κ as an unknown calibration constant of unspecified magnitude, and the experiment proposed in Section 4 would estimate it by comparing the index with measured strain. The index is useful for ranking exposures within one implant design and bone type, and unreliable for comparing absolute strains with historical thresholds. The moment that corresponds to a chosen strain ε is
M_ε=(ε E D^3)/κ
Illustrative computation under stated assumptions — not empirical data. The script saved as overload_illustration.py in the paper folder was run in this session. Table 2 gives moments M = F_lat·h for forces between 20 and 795 N and lever arms between 5 and 20 mm. The force values 434 N and 795 N are the mean and peak in vivo canine bite loads from Kan and colleagues [8]; the lever arms are assumed values chosen to span a short crown to a cantilever, and the lateral component and direction are unknown for that study.
| F_lat (N) | h = 5 mm | h = 10 mm | h = 15 mm | h = 20 mm |
|---|---|---|---|---|
| 20 | 10.0 | 20.0 | 30.0 | 40.0 |
| 50 | 25.0 | 50.0 | 75.0 | 100.0 |
| 100 | 50.0 | 100.0 | 150.0 | 200.0 |
| 200 | 100.0 | 200.0 | 300.0 | 400.0 |
| 434 | 217.0 | 434.0 | 651.0 | 868.0 |
| 795 | 397.5 | 795.0 | 1192.5 | 1590.0 |
The table shows how widely the moment can vary for a given force: a fourfold change in lever arm changes the moment fourfold, which is the same order as the difference between a 100 N and a 434 N lateral force. Rows 434 and 795 treat the whole measured load as lateral, which is an upper bound; for a 30-degree inclination the script computed 217.0 and 325.5 N·cm for 434 N at 10 and 15 mm.
For comparison with the animal data of Duyck and colleagues, the script recomputed their reported moments from their reported forces and arms: 29.4 N at 1.5 mm gave 4.41 N·cm, and 14.7 N at 50 mm gave 73.5 N·cm, in agreement with the values in their abstract [20].
For the strain index with E = 15,000 N/mm² (an assumed value representing cortical bone) and D = 4.0 mm, the script computed the moment corresponding to 3000 microstrain as 28.3 N·cm for κ = κ₀, 56.5 N·cm for κ = κ₀/2, and 113.1 N·cm for κ = κ₀/4. For D = 3.5 mm the values were 18.9, 37.9, and 75.8 N·cm, and for D = 5.0 mm they were 55.2, 110.4, and 220.9 N·cm. Two points follow. A fourfold uncertainty in κ gives a fourfold uncertainty in the moment at which any strain threshold is reached, so a strain threshold cannot be translated into a force without calibration. Also, M₃₀₀₀ scales with D³: widening the implant from 3.5 to 5.0 mm raises the moment needed for the same index strain about 2.9-fold (55.2 divided by 18.9 for κ = κ₀). A moment threshold M* that is identical across diameters is therefore not expected if strain is the controlling variable, and Prediction P5 below tests that.
The cycle dimension
A static threshold in M is incomplete. Bone is a material that accumulates damage under repeated loading, and the fatigue literature in orthopaedics is built on strain amplitude and cycle number together. We propose only the functional form. Let loss occur when
M>M^*(N)=M_∞+A N^(−b)
where M∞ is a moment tolerated indefinitely, and A (N·cm) and b (dimensionless, positive) describe how much extra moment is tolerated for a short exposure. All three are unknown and species and site specific. The form has two testable consequences. The first is that very high moments produce loss after few cycles, which the dynamic rabbit protocol, with 2,520 cycles in 14 days [20], is consistent with in a qualitative way but cannot confirm. The second is that moderate moments produce loss only after long exposure, which would explain why Isidor’s implants failed after 4.5 to 15.5 months rather than quickly [1].
As an order-of-magnitude illustration, the same script computed loaded-cycle counts for an assumed rate of 1,000 or 2,000 loaded cycles per day. Those are placeholder rates that do not come from any source. Eight months (240 days) gave 240,000 to 480,000 cycles; 4.5 months gave 135,000 to 270,000; and 15.5 months gave 465,000 to 930,000. By comparison the rabbit protocol applied 2,520 cycles [20]. Cycle counts in chronic oral studies therefore exceed those of the only quantified dynamic study by roughly two orders of magnitude if every chewing cycle counted at the overload moment, which is an upper bound. Most chewing cycles are not at peak force, and the fraction of cycles at near-peak load is unknown in all the animal models discussed here.
Placing the published models on one scale
Table 3 lists, for each model, what an abstract-level reading allows and what remains unknown. The point of the table is not to compute moments for these studies, which would require invented inputs. It is to show that the retrospective re-expression proposed in Section 4 depends on quantities that most reports do not state.
| Model | Force F | Lever arm h or offset | Direction | Cycles and duration | Inflammation control |
|---|---|---|---|---|---|
| Isidor 1996, monkeys [1] | Unknown | Unknown (two-implant prosthesis; geometry not stated) | Lateral after prosthesis renewal (stated) | Not counted; loss at 4.5–15.5 months | Weekly brushing and monthly subgingival cleaning on loaded implants |
| Miyata 2000, monkeys [17] | Unknown; excess height 100, 180, 250 micrometers | Unknown | Unknown | Not stated in abstract | Kept inflammation-free (stated) |
| Heitz-Mayfield 2004, dogs [2] | Unknown | Unknown (crown height not stated) | Unknown | 8 months loaded; not counted | Plaque control throughout |
| Kozlovsky 2007, dogs [3] | Unknown | Unknown | Unknown | 12 months; not counted | Brushing three times weekly in the uninflamed arm |
| Lima 2019, dogs [4] | Unknown | Cantilever length not stated | Unknown | 24 weeks of radiographs; not counted | Bleeding on probing recorded |
| Duyck 2001, rabbit tibia [20] | 29.4 N static; 14.7 N dynamic | 1.5 mm; 50 mm | Transverse | 2,520 cycles at 1 Hz over 14 days | Not applicable to the abstract’s bicortical tibial model |
| Kan 2014, dogs [8] | Mean 434 N; peak 795 N (device measured) | Unknown | Unknown | 8 weeks | Plaque control prescribed |
Numbered predictions and refuting results
Each prediction is stated with the result that would refute it. All presuppose inflammation control (I = 0) unless stated.
P1. Threshold form. In the dose–response experiment of Section 4, the probability of loss of osseointegration at fixed N is near zero at low M and rises sharply over a narrow range, and a segmented (change-point) model fits significantly better than a log-linear dose model. Refuted if a smooth log-linear or logistic model fits as well or better and the confidence interval of any change point spans most of the dose range.
P2. Sub-threshold neutrality or gain. Below M*, implants show bone-to-implant contact and crestal bone level no worse than unloaded controls, with possible gain, as in some dog studies [3, 4]. Refuted if crestal bone loss increases monotonically from the lowest loaded level.
P3. Cycle dependence. At a moment above M∞, time to loss shortens as N accumulates faster, and M*(N) is non-increasing. Refuted if outcomes at equal M do not depend on N over a range of at least tenfold, or if outcomes depend only on N.
P4. Retrospective ordering. When full-text inputs allow estimating M (with intervals) for the published models, the Isidor model has the highest estimated moment and the null dog models lower ones, with the intervals not overlapping the threshold in the wrong direction. Refuted if a null model has an interval of M wholly above that of the positive model, or if the intervals are so wide that every ordering is possible (in which case the retrospective test is uninformative, which would itself be reported).
P5. Geometry scaling. If strain is the controlling variable, M* rises with implant diameter roughly as D³ (a ratio of about 2.9 between 3.5 and 5.0 mm diameters under κ = κ₀). Refuted if M* is statistically indistinguishable between diameters while measured crestal strain differs by the predicted factor, in which case moment, not strain, is the relevant variable, or if neither variable orders outcomes.
P6. Strain threshold not at 3000 microstrain. Back-calculated crestal strain at sites that resorbed does not cluster at or above 3000 microstrain; the strain at which risk rises depends on the preparation. Refuted if resorbing sites cluster at or above 3000 microstrain and non-resorbing sites below it across models.
P7. Inflammation interaction. The same moment produces more bone loss in inflamed than in uninflamed sites, so that M*_inflamed is below M*. Refuted if the difference in loss between loaded and unloaded implants is the same in inflamed and uninflamed arms. The Kozlovsky result is consistent with this prediction but its sample was four animals [3].
P8. Direction-independence given moment. At equal M, outcomes do not differ systematically between loading with a large lever arm and small force and one with a small lever arm and large force. Refuted if outcomes differ at equal M, which would mean that force direction or the axial component carries independent information.
4. Proposed methodology
Overview
The proposal has two components. Component A is a retrospective, full-text extraction of force, geometry, direction, cycles, and inflammation control from the published overload models and, where authors agree, from their archived records. Component B is a prospective large-animal experiment with controlled and measured loading. Neither has been performed. Component A would be cheap and could be completed first. Component B requires institutional animal-ethics approval, which has not been sought, and registration of the protocol before the first implant is placed; neither is claimed here.
Component A: retrospective re-expression
Two reviewers, working independently, would extract from each full text the implant diameter and length, crown or cantilever height from the bone crest, antagonist type, occlusal contact registration, any force or bite-force measurement, loading duration, and inflammation status. Where force was not measured, species-specific bite-force data would be used as an interval, for example measured forces in dogs [8, 24], and the resulting M intervals would be propagated using stated distributions. The output would be a table of M and N intervals with the reported outcome. Because there are perhaps eight to twelve eligible studies, no inferential claim is planned beyond rank ordering and a description of overlap. The component tests P4, and it also shows which studies are uninformative because their intervals span the whole range. Micro-CT or histology archives from published models, where available, would allow finite-element models to estimate strain at sites that did and did not resorb, which would test P6.
Component B: prospective dose–response experiment
Species and sites. A large animal with a mandible that accepts standard-diameter implants in healed alveolar ridges (dog or minipig) is appropriate. Implants of one design, a single diameter (illustratively 4.0 mm) and one surface would be used in the primary analysis; a second diameter would be used in a pre-specified secondary analysis for P5.
Exposure. The central requirement is that dose is imposed and measured rather than assumed. Abutments would carry a bending-moment sensor (strain-gauge bridge) so that the moment of each cycle can be logged. The load would be applied by an intraoral loader at a fixed lever arm with a force controller, with a verified cycle counter, so that M and N are controlled and the loader’s force is checked by the sensor. The dose levels are on a nominal scale of 0 (instrumented but unloaded), 25, 50, 100, and 200 N·cm, chosen to bracket the range of M₃₀₀₀ values computed in Section 3.3 for a 4.0 mm implant under the assumed κ range (28.3 to 113.1 N·cm). These levels would be revised after a pilot in which κ is estimated from strain measured on the animal’s bone. Each animal would receive one implant per dose level, positions randomized by a Latin-square scheme.
Cycles. The primary analysis would use a single, pre-specified, long exposure at the target moment. Secondary cohorts, killed at earlier time points, would give N dependence for P3, at the cost of more animals; a single-cohort design with interval-censored mobility testing is the economical alternative.
Inflammation control. Daily mechanical plaque control and professional cleaning would be used, with bleeding on probing, a mucosal index, and histologic inflammatory infiltrate recorded for each implant as covariates. A separate, optional ligature-induced inflammation stratum would test P7.
Outcomes. The primary outcome is a composite of loss of osseointegration, defined before the first loading as any of: clinical mobility, continuous peri-implant radiolucency on micro-CT, or bone-to-implant contact below a pre-set fraction of the animal’s unloaded control. Secondary outcomes are crestal bone level (micro-CT), bone-to-implant contact (histomorphometry), resonance frequency analysis, and crestal strain from gauges or digital image correlation on explanted specimens loaded to calibration. Assessors of micro-CT and histology would be blinded to dose.
Sample size reasoning
Sample-size planning depends on a dose–response shape that is unknown, so we computed power in two scenarios by Monte Carlo simulation. Illustrative computation under stated assumptions — not empirical data. Parameters: five dose levels (0, 25, 50, 100, 200 N·cm), one implant per level per animal, an animal-level random intercept on the logit scale with standard deviation 0 or 1, a logistic dose model on the scale log(1 + dose/10), 2,000 simulated experiments per cell, random seed 24, and a likelihood-ratio test for a positive slope at a nominal two-sided level of 0.05 (the analysis ignored clustering, which is a simplification). The scenarios assume probabilities of loss at the five dose levels, not data.
Steep scenario (probabilities 0.02, 0.05, 0.20, 0.60, 0.90). With a random-intercept standard deviation of 0 the computed power was 0.971 for 4 animals (20 implants), 0.997 for 6, and 1.000 for 8 or more. With a standard deviation of 1.0 the computed power was 0.951 for 4 animals, 0.994 for 6, 0.998 for 8, and 1.000 for 10 and 12.
Shallow scenario (probabilities 0.05, 0.08, 0.12, 0.20, 0.30; standard deviation 1.0). The computed power was 0.376 for 8 animals (40 implants), 0.500 for 12, 0.605 for 16, 0.763 for 24, and 0.860 for 32.
If the true curve is steep, detecting a dose effect needs few animals; the simulation does not address localizing a threshold or distinguishing it from a smooth curve, which needs more dose levels near the transition. We therefore propose a staged design: a pilot of about six animals to estimate the slope and κ, followed by a pre-specified sample-size re-estimate using the formula-free simulation above refitted to pilot data. The precision of the threshold estimate was not simulated here and should be before submission.
Statistical analysis plan
The primary model is a mixed-effects logistic regression of the composite outcome on log(1 + M/10) with an animal random intercept, compared with a segmented (change-point) version; the comparison uses a likelihood-ratio statistic with a parametric bootstrap reference distribution, because the change point is not a regular parameter. The threshold M* is reported with a bootstrap interval. For P3 a time-to-event or interval-censored model includes M, N (cumulative cycles), and their interaction. P1 is the single primary hypothesis; all others are secondary and hypothesis-generating. Early implant losses before loading are excluded and reported. A hold-out sample of animals would test the out-of-sample prediction of the estimated M*, which is the main protection against post hoc threshold selection.
Bias control, ethics and feasibility
Dose assignment would be randomized and concealed from the surgeon and assessors, sensor calibration would be checked before and after the experiment, implants with sensor failure would be reported, and the study would follow the ARRIVE reporting guidelines. The feasibility risks are the cost and size of instrumented abutments in a large-animal mouth, drift in the gauges over months, and loader-related soft-tissue trauma that would confound inflammation. No ethics approval, registration, or funding exists for this proposal.
5. Competing explanations and how the design discriminates among them
Graded risk without a threshold. Risk of loss may rise smoothly with dose, so that a “threshold” is a convenience of a steep part of a curve. The design discriminates by including dose levels on both sides of any plausible transition and by comparing a segmented model with a smooth model using a bootstrap reference distribution (Section 4.5). If the smooth model fits equally, the threshold language should be dropped and the dose-response curve reported as such.
Cumulative damage rather than a threshold in M and N. If damage accumulates as a power of the moment ratio, outcomes depend on the integral over the load history. The design discriminates only if the exposure includes variable-amplitude blocks (for instance, equal total damage under different mixes of high and low moments). We list this as an optional arm; without it, the design cannot separate a threshold surface M*(N) from cumulative damage.
Direction and shear rather than magnitude of moment. Isidor’s model became lateral after prosthesis renewal [1]. If the injurious variable is transverse shear at the crest or the axial component of force, moment will not order outcomes. The design uses a fixed lever arm, but P8 deliberately varies the force and the arm at equal M so that moment and force separate.
Occult inflammation or micromotion. Loss of osseointegration under load may be mediated by a subclinical inflammatory process or by micromotion at the interface, which is known to compromise osseointegration [6]. A micromotion tolerance range of tens of micrometers has been derived from the literature [36], but it is a literature-derived range, not a measured threshold for crestal bone. The design measures mucosal indices, histologic inflammatory infiltrate, and resonance frequency, so that loss with histologic inflammation can be separated from loss without. If most overload losses coincide with histologic inflammation, the hypothesis is not refuted, but its scope shrinks to a modifier effect (P7).
Stochastic loss. Loss of osseointegration may be a rare, near-random event unrelated to dose at accessible magnitudes. P1 and P2 would then fail, and the correct conclusion would be that overload alone is not a practical cause in inflammation-free sites, which is the reading of Duyck and Vandamme [6].
6. Clinical and engineering implications
None of the following is a recommendation. They are conditional on the hypothesis surviving the experiments of Section 4.
If a moment threshold exists and is reproducible across species when scaled by geometry, clinicians would have a reason to think in terms of lever arm as well as occlusal contact. A cantilever or tall crown increases the moment for a given force, whereas a large-diameter implant raises the strain tolerance of the interface according to the D³ scaling of the index. For engineering, the more useful consequence is a measurable quantity at the abutment. Instrumented or sensor-equipped abutments exist in the research literature [22, 27], and a calibrated moment would allow testing of overload claims, which the proxies reviewed in Section 2.5 cannot do.
For simulation, a validated threshold in moment or strain would replace an imported value. Until then, remodeling models that include an overload term [12–14] should report sensitivity to the threshold they adopt, because the evidence retrieved shows that the only direct confrontation of the value with measured bite load and strain in an overload model found strains well below it [8].
7. Limitations
- Abstract-level appraisal. Full texts, which may contain forces, crown heights, and occlusal records, were not read. Statements that a study “did not report” a quantity mean that the abstract did not.
- Search scope. PubMed and Europe PMC were searched; Embase, Scopus, and Web of Science were not accessible. Some quantified studies may have been missed, and the originality assessment in the Novelty section is correspondingly provisional.
- The strain index is a scaling device. κ is unknown, a cylinder in bone is not a free cantilever, and the cortical modulus used in the illustration is an assumption. The illustrative moments in Table 2 use assumed forces and lever arms and describe no real experiment.
- The Monte Carlo illustration. It assumes the dose-response shape, ignores clustering in the analysis, uses a nominal likelihood-ratio test, and does not simulate threshold localization.
- Human relevance. Humans are exposed to clinical risk factors (smoking, hygiene) that dominate clinical series [33], and an animal threshold would not translate into a clinical one without further work.
- No data exist. The hypothesis is untested, and its main prediction may fail.
8. Conclusions
The animal literature on occlusal overload cannot currently be placed on one dose scale because most abstracts do not state force, lever arm, direction, or cycle number. Within that limit, a threshold hypothesis in crestal moment and cycle number, absent inflammation, is a coherent and falsifiable reading of the conflict between the monkey and dog models, but no evidence reviewed here establishes it. The 3000-microstrain figure is an extrapolation from long-bone theory that the one direct test we found did not reach in a canine overload preparation, and it has no validation for peri-implant bone. The decisive step is an instrumented dose–response experiment with a pre-specified analysis, which could also show that no threshold exists.
Cite this paper
Jain S, Gupta A. A Load-Quantified Threshold Hypothesis for Loss of Osseointegration Under Occlusal Overload. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/occlusal-overload-threshold
References
- Isidor F. Loss of osseointegration caused by occlusal load of oral implants. A clinical and radiographic study in monkeys. Clin Oral Implants Res. 1996;7(2):143-52. doi:10.1034/j.1600-0501.1996.070208.x (opens in a new tab). PMID: 9002833 (opens in a new tab).
- Heitz-Mayfield LJ, Schmid B, Weigel C, Gerber S, Bosshardt DD, Jönsson J, et al. Does excessive occlusal load affect osseointegration? An experimental study in the dog. Clin Oral Implants Res. 2004;15(3):259-68. doi:10.1111/j.1600-0501.2004.01019.x (opens in a new tab). PMID: 15142087 (opens in a new tab).
- Kozlovsky A, Tal H, Laufer BZ, Leshem R, Rohrer MD, Weinreb M, et al. Impact of implant overloading on the peri-implant bone in inflamed and non-inflamed peri-implant mucosa. Clin Oral Implants Res. 2007;18(5):601-10. doi:10.1111/j.1600-0501.2007.01374.x (opens in a new tab). PMID: 17655715 (opens in a new tab).
- Lima LA, Bosshardt DD, Chambrone L, Araújo MG, Lang NP. Excessive occlusal load on chemically modified and moderately rough titanium implants restored with cantilever reconstructions. An experimental study in dogs. Clin Oral Implants Res. 2019;30(11):1142-1154. doi:10.1111/clr.13539 (opens in a new tab). PMID: 31529643 (opens in a new tab).
- Naert I, Duyck J, Vandamme K. Occlusal overload and bone/implant loss. Clin Oral Implants Res. 2012;23 Suppl 6:95-107. doi:10.1111/j.1600-0501.2012.02550.x (opens in a new tab). PMID: 23062133 (opens in a new tab).
- Duyck J, Vandamme K. The effect of loading on peri-implant bone: a critical review of the literature. J Oral Rehabil. 2014;41(10):783-94. doi:10.1111/joor.12195 (opens in a new tab). PMID: 24889500 (opens in a new tab).
- Frost HM. Bone’s mechanostat: a 2003 update. Anat Rec A Discov Mol Cell Evol Biol. 2003;275(2):1081-101. doi:10.1002/ar.a.10119 (opens in a new tab). PMID: 14613308 (opens in a new tab).
- Kan JP, Judge RB, Palamara JE. In vitro bone strain analysis of implant following occlusal overload. Clin Oral Implants Res. 2014;25(2):e73-82. doi:10.1111/clr.12059 (opens in a new tab). PMID: 23067316 (opens in a new tab).
- Frost HM. A 2003 update of bone physiology and Wolff’s Law for clinicians. Angle Orthod. 2004;74(1):3-15. doi:10.1043/0003-3219(2004)074<0003:AUOBPA>2.0.CO;2 (opens in a new tab). PMID: 15038485 (opens in a new tab).
- Stanford CM, Brand RA. Toward an understanding of implant occlusion and strain adaptive bone modeling and remodeling. J Prosthet Dent. 1999;81(5):553-61. doi:10.1016/s0022-3913(99)70209-x (opens in a new tab). PMID: 10220659 (opens in a new tab).
- Kuc AE, Sulewska M, Hajduk G, Kuc P, Kuc N, Lis J, et al. Osteocyte Mechanobiology in Peri-Implant Bone Adaptation: A Narrative Review and Hypothesis-Generating Framework for SOST/Wnt-Linked Cortical Stability. Cells. 2026;15(9). doi:10.3390/cells15090748 (opens in a new tab). PMID: 42121850 (opens in a new tab).
- Li J, Li H, Shi L, Fok AS, Ucer C, Devlin H, et al. A mathematical model for simulating the bone remodeling process under mechanical stimulus. Dent Mater. 2007;23(9):1073-8. doi:10.1016/j.dental.2006.10.004 (opens in a new tab). PMID: 17137621 (opens in a new tab).
- Chou HY, Jagodnik JJ, Müftü S. Predictions of bone remodeling around dental implant systems. J Biomech. 2008;41(6):1365-73. doi:10.1016/j.jbiomech.2008.01.032 (opens in a new tab). PMID: 18394631 (opens in a new tab).
- Eser A, Tonuk E, Akca K, Dard MM, Cehreli MC. Predicting bone remodeling around tissue- and bone-level dental implants used in reduced bone width. J Biomech. 2013;46(13):2250-7. doi:10.1016/j.jbiomech.2013.06.025 (opens in a new tab). PMID: 23876712 (opens in a new tab).
- Mellal A, Wiskott HW, Botsis J, Scherrer SS, Belser UC. Stimulating effect of implant loading on surrounding bone. Comparison of three numerical models and validation by in vivo data. Clin Oral Implants Res. 2004;15(2):239-48. doi:10.1111/j.1600-0501.2004.01000.x (opens in a new tab). PMID: 15008937 (opens in a new tab).
- Piccinini M, Cugnoni J, Botsis J, Ammann P, Wiskott A. Numerical prediction of peri-implant bone adaptation: Comparison of mechanical stimuli and sensitivity to modeling parameters. Med Eng Phys. 2016;38(11):1348-1359. doi:10.1016/j.medengphy.2016.08.008 (opens in a new tab). PMID: 27641659 (opens in a new tab).
- Miyata T, Kobayashi Y, Araki H, Ohto T, Shin K. The influence of controlled occlusal overload on peri-implant tissue. Part 3: A histologic study in monkeys. Int J Oral Maxillofac Implants. 2000;15(3):425-31. PMID: 10874809 (opens in a new tab).
- Gotfredsen K, Berglundh T, Lindhe J. Bone reactions at implants subjected to experimental peri-implantitis and static load. A study in the dog. J Clin Periodontol. 2002;29(2):144-51. doi:10.1034/j.1600-051x.2002.290209.x (opens in a new tab). PMID: 11895542 (opens in a new tab).
- Podaropoulos L, Veis AA, Trisi P, Papadimitriou S, Alexandridis C, Kalyvas D. Bone reactions around dental implants subjected to progressive static load: an experimental study in dogs. Clin Oral Implants Res. 2016;27(7):910-7. doi:10.1111/clr.12658 (opens in a new tab). PMID: 26202378 (opens in a new tab).
- Duyck J, Rønold HJ, Van Oosterwyck H, Naert I, Vander Sloten J, Ellingsen JE. The influence of static and dynamic loading on marginal bone reactions around osseointegrated implants: an animal experimental study. Clin Oral Implants Res. 2001;12(3):207-18. doi:10.1034/j.1600-0501.2001.012003207.x (opens in a new tab). PMID: 11359477 (opens in a new tab).
- Nagasawa M, Takano R, Maeda T, Uoshima K. Observation of the bone surrounding an overloaded implant in a novel rat model. Int J Oral Maxillofac Implants. 2013;28(1):109-16. doi:10.11607/jomi.2388 (opens in a new tab). PMID: 23377055 (opens in a new tab).
- Assoratgoon I, Yoda N, Iwamoto M, Sato T, Kawata T, Egusa H, et al. In vivo measurement of three-dimensional load exerted on dental implants: a literature review. Int J Implant Dent. 2022;8(1):52. doi:10.1186/s40729-022-00454-y (opens in a new tab). PMID: 36372828 (opens in a new tab).
- Duyck J, Van Oosterwyck H, Vander Sloten J, De Cooman M, Puers R, Naert I. Magnitude and distribution of occlusal forces on oral implants supporting fixed prostheses: an in vivo study. Clin Oral Implants Res. 2000;11(5):465-75. doi:10.1034/j.1600-0501.2000.011005465.x (opens in a new tab). PMID: 11168239 (opens in a new tab).
- Kim SE, Arzi B, Garcia TC, Verstraete FJM. Bite Forces and Their Measurement in Dogs and Cats. Front Vet Sci. 2018;5:76. doi:10.3389/fvets.2018.00076 (opens in a new tab). PMID: 29755988 (opens in a new tab).
- Hylander WL. In-vivo bone strain as an indicator of masticatory bite force in Macaca fascicularis. Arch Oral Biol. 1986;31(3):149-57. doi:10.1016/0003-9969(86)90120-2 (opens in a new tab). PMID: 3459403 (opens in a new tab).
- Panagiotopoulou O, Iriarte-Diaz J, Wilshin S, Dechow PC, Taylor AB, Mehari Abraha H, et al. In vivo bone strain and finite element modeling of a rhesus macaque mandible during mastication. Zoology (Jena). 2017;124:13-29. doi:10.1016/j.zool.2017.08.010 (opens in a new tab). PMID: 29037463 (opens in a new tab).
- Cozzolino F, Apicella D, Wang G, Apicella A, Sorrentino R. Implant-to-bone force transmission: a pilot study for in vivo strain gauge measurement technique. J Mech Behav Biomed Mater. 2019;90:173-181. doi:10.1016/j.jmbbm.2018.10.014 (opens in a new tab). PMID: 30366308 (opens in a new tab).
- Mojaver S, Patel N, Sarmiento H, Fiorellini JP. Under pressure: Unraveling the impact of occlusal overload on peri-implant health-A systematic review. J Prosthodont. 2025;34(8):784-795. doi:10.1111/jopr.14088 (opens in a new tab). PMID: 40571905 (opens in a new tab).
- Di Fiore A, Montagner M, Sivolella S, Stellini E, Yilmaz B, Brunello G. Peri-Implant Bone Loss and Overload: A Systematic Review Focusing on Occlusal Analysis through Digital and Analogic Methods. J Clin Med. 2022;11(16). doi:10.3390/jcm11164812 (opens in a new tab). PMID: 36013048 (opens in a new tab).
- Bertolini MM, Del Bel Cury AA, Pizzoloto L, Acapa IRH, Shibli JA, Bordin D. Does traumatic occlusal forces lead to peri-implant bone loss? A systematic review. Braz Oral Res. 2019;33(suppl 1):e069. doi:10.1590/1807-3107bor-2019.vol33.0069 (opens in a new tab). PMID: 31576953 (opens in a new tab).
- Dalago HR, Schuldt Filho G, Rodrigues MA, Renvert S, Bianchini MA. Risk indicators for Peri-implantitis. A cross-sectional study with 916 implants. Clin Oral Implants Res. 2017;28(2):144-150. doi:10.1111/clr.12772 (opens in a new tab). PMID: 26754342 (opens in a new tab).
- Engel E, Gomez-Roman G, Axmann-Krcmar D. Effect of occlusal wear on bone loss and Periotest value of dental implants. Int J Prosthodont. 2001;14(5):444-50. PMID: 12066640 (opens in a new tab).
- Lindquist LW, Carlsson GE, Jemt T. A prospective 15-year follow-up study of mandibular fixed prostheses supported by osseointegrated implants. Clinical results and marginal bone loss. Clin Oral Implants Res. 1996;7(4):329-36. doi:10.1034/j.1600-0501.1996.070405.x (opens in a new tab). PMID: 9151599 (opens in a new tab).
- Schwarz F, Derks J, Monje A, Wang HL. Peri-implantitis. J Clin Periodontol. 2018;45 Suppl 20:S246-S266. doi:10.1111/jcpe.12954 (opens in a new tab). PMID: 29926484 (opens in a new tab).
- Afrashtehfar KI, Afrashtehfar CD. Lack of association between overload and peri-implant tissue loss in healthy conditions. Evid Based Dent. 2016;17(3):92-93. doi:10.1038/sj.ebd.6401193 (opens in a new tab). PMID: 27767107 (opens in a new tab).
- Szmukler-Moncler S, Salama H, Reingewirtz Y, Dubruille JH. Timing of loading and effect of micromotion on bone-dental implant interface: review of experimental literature. J Biomed Mater Res. 1998;43(2):192-203. doi:10.1002/(sici)1097-4636(199822)43:2<192::aid-jbm14>3.0.co;2-k (opens in a new tab). PMID: 9619438 (opens in a new tab).
- Chambrone L, Chambrone LA, Lima LA. Effects of occlusal overload on peri-implant tissue health: a systematic review of animal-model studies. J Periodontol. 2010;81(10):1367-78. doi:10.1902/jop.2010.100176 (opens in a new tab). PMID: 20507230 (opens in a new tab).

