Skip to main content

What Does the Implant Stability Quotient Measure? A Mechanical Model of Resonance Frequency Analysis and Its Consequences for Loading Thresholds

Paper type
Methodological study (theoretical)
Status
Original Work · Working Paper
Keywords:
implant stability quotient · resonance frequency analysis · insertion torque · primary stability · measurement error · lumped mechanical model · loading thresholds
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. Placement-time insertion torque (about 35 N·cm) and implant stability quotient (ISQ, about 60 to 70) are widely used to decide when an implant may be loaded. Pooled analyses report only a moderate ISQ-torque correlation, no consistent relation of ISQ to bone-implant contact, and near-chance discrimination of later failure. The 2 to 4 week “stability dip” is taught as a biological event, yet its reported size is a few ISQ units.

Hypothesis. ISQ is the proprietary rescaling of the first bending resonance of a transducer-implant-bone system. At placement it is governed mainly by the exposed (unsupported) length of the system and the stiffness of the crestal cortex, with a smaller trabecular contribution. Insertion torque integrates frictional and compressive resistance along the whole thread length. The two therefore share variance but measure different mechanical quantities, and neither isolates the interface property most plausibly linked to failure. The reported dip is of the same order as the smallest detectable change of an individual reading.

Proposed approach. We specify a lumped cantilever-on-rotational-spring model with defined symbols, report an illustrative sensitivity calculation and Monte Carlo simulation under stated assumptions (not empirical data), derive smallest-detectable-change arguments from published reliability figures, state eight falsifiable predictions, and outline a four-phase validation program: factorial bench testing, cadaver variance partitioning, repeated-measures reliability with phantom controls, and individual-participant reanalysis.

Expected contribution. The mechanical dependence of resonance frequency on cortical thickness and exposed height is already established in finite element, in vitro and clinical work. The contribution is therefore narrowed to three items: a compact explicit model, a conditional account of how construct mismatch can depress prognostic validity, and an individual-versus-group measurement-error framework for the dip.

Conclusions. Under the assumptions made, ISQ is more informative about geometry and crestal stiffness than about the interface property that matters for survival, and an individual 3 to 4 unit decline is not interpretable without device-specific reliability data. The model requires bench calibration before any clinical inference.

1. Introduction

Two numbers dominate decisions about early loading of dental implants. The first is the peak torque at which the implant seated, with 35 N·cm commonly cited as a minimum for immediate loading. The second is the implant stability quotient, a dimensionless index displayed by resonance frequency analysis (RFA) devices, with values of about 60 to 70 often treated as a green zone. Both are used as if they quantified how firmly bone holds the implant. If they quantify different things, a clinician who treats them as interchangeable surrogates for “primary stability” will be misled in a systematic way.

The evidence base shows several inconsistencies that a single mechanical account might explain. A meta-analysis of 48 studies reported a pooled ISQ-torque correlation of 0.44 with very high heterogeneity, and found no significant link between RFA values and survival [1]. An earlier systematic review of 12 studies reported a Spearman coefficient of 0.366 that did not reach significance [2]. In beagle dogs, ISQ did not correlate with histomorphometric bone-implant contact [3]. In human cadavers, ISQ was related to cortical thickness but not to the other bone structure variables examined [4]. Diagnostic accuracy of RFA for failure of immediately loaded implants was close to chance in one meta-analysis [5], and recent meta-analyses did not find survival differences across the torque or ISQ thresholds [6, 7]. Meanwhile, the ISQ decline at 2 to 4 weeks, reported as 3 to 4 units in one prospective series [8], is interpreted in teaching as the handover from mechanical to biological stability.

This manuscript asks three questions. (i) What mechanical quantity does a resonance reading actually encode, and how does that differ from the quantity encoded by insertion torque? (ii) If the two differ, how large an effect would that difference have on the ability of placement thresholds to discriminate failures? (iii) Is the early dip distinguishable from measurement error at the level of a single implant?

Before building anything, we assumed that the mechanical interpretation had already been published and searched for it. It largely has (Section 2.3). We therefore do not present the mechanical dependence of ISQ on cortical thickness and exposed height as new. The manuscript is a methodological study of theory: it assembles a transparent model, tests its internal consistency by computation under stated assumptions, derives what it implies for prognostic validity and measurement error, and specifies how it could be refuted. No experiment or patient-data analysis was performed, and every computed number below comes from code saved with the manuscript.

2. Background and critical appraisal of the evidence

Origin of the concept

Resonance frequency analysis was introduced for dental implants as a way to quantify the stability of the implant-tissue interface through the frequency response of a small transducer attached to the implant [9]. Subsequent animal work followed the frequency of titanium implants during healing in the rabbit tibia [10], and cadaver work compared the initial stability of different implant designs [11]. A review by Sennerby and Meredith summarized biological and biomechanical aspects of RFA and its clinical implications [12]. The physical reading in these early papers was that frequency rises with the stiffness of the support and falls with the free length of the vibrating system. The commercial ISQ scale converts frequency into an index; the exact transformation is proprietary, and details are not in the sources available to us. We therefore treat ISQ only as a monotonically increasing function of frequency and make no claim about its local slope.

Evidence that ISQ and torque measure different things

Table 1 lists the pivotal studies with design, size, main bias and an operational evidence level.

Table 1. Pivotal studies bearing on the hypothesis, with design, size, principal bias and operational evidence level.
StudyDesign and sizeMain finding (abstract level)Main bias or limitationLevel
Tisci 2026 [1]Meta-analysis, 48 studiesISQ-torque r 0.44 (95% CI 0.32-0.55), I2 >90%; no significant link to survivalHeterogeneous devices, bone types and implant designsB
Lages 2018 [2]Systematic review, 12 studiesrs 0.366, not significantSmall number of studies, mixed methodsB
Manresa 2014 [3]Dog, 36 implants, 6 animalsNo ISQ-BIC correlationOne species, one time pointC
Rozé 2009 [4]Human cadaver, 22 implantsISQ related only to cortical thicknessSmall n, ex vivoC
Hsu 2013 [13]Synthetic bone, 60 specimensISQ, torque and Periotest all depend on cortex and trabecular modulusIdealized materials, no biologyC
Winter 2010 [14]Finite elementRFA rises with cortical thickness and length, falls with bone lossAssumed material propertiesD
Lages 2017 [15]Cross-sectional, 31 implantsISQ falls as transmucosal height risesSmall sample, single designC
Atieh 2012 [5]Diagnostic meta-analysis, 15 studies, 2,236 implantsAUC 0.54Heterogeneous thresholds and outcomesB
Razaghi 2025 [7]Meta-analysis, 79 studiesNo threshold effect for ISQ 60 or torque 35Aggregate data, low failure countsB
Atieh 2014 [16]Cohort, 28 implantsISQ at 8 weeks discriminated well (AUC 0.93)Tiny n, optimismC
Trisi 2016 [17]Sheep, after healingMicromotion correlated with ISQ and BICHealed interfaces, animal modelC
Norton 2019 [18]210 implants, three instruments82% of Penguin-Osstell pairs within 4 unitsSingle centerC

Note. Evidence levels are operational categories used in this manuscript, not GRADE: A strong, B moderate, C preliminary, D mechanistic or theoretical, E speculative.

The cortex-dominance reading is supported by several independent designs. In a cadaver study, ISQ was related to cortical thickness alone [4]. A clinical series using CBCT found a strong relation between crestal cortical thickness and ISQ, with the relation depending on measurement depth, being strongest within about 3 mm of the crest [19]. Another clinical series found only a weak positive correlation between cortical thickness and ISQ at placement, with differences by jaw and by augmentation status [20]. A prospective clinical, biomechanical and imaging study examined cortical thickness and implant length in relation to stability at surgery [21]. In synthetic bone, cortex thickness explained most of the variance of ISQ, but so did trabecular modulus, so cortex dominance is not exclusive [13].

Exposed height has equally direct evidence. The clearest is a within-patient comparison in which the same implants were measured at platform level and on abutments of two transmucosal heights, with mean ISQ values of 88.27, 72.75 and 66.33, respectively [15]. A prospective clinical study of healing abutment height reported higher secondary stability with taller abutments placed closer to the antagonist [22], which appears to run opposite to the cantilever logic, but occlusal contact is a plausible confounder that the abstract does not address. Finite element work likewise predicts that bone loss lowers RFA and that cortical thickness and length raise it, with nonlinear responses to linear changes in length and bone loss [14].

Insertion torque differs mechanically. It accumulates over the full threaded length as bone is compressed and cut, and it depends on drilling protocol, thread geometry and the time-history of seating. The Hsu synthetic bone study found that torque, Periotest and ISQ all responded to cortex thickness and trabecular modulus [13], so torque also contains a cortical component. The claim is not that torque is cortex-free, only that the weighting differs. In a systematic review, torque but not ISQ was related to marginal bone loss, while neither predicted survival and primary and secondary stability were strongly related [23].

Originality challenge: what has already been done

Under three vocabularies (finite element; in vitro simulated bone; analytical and vibration models) and in Europe PMC preprints, the following were found. Finite element models of RFA include a parametric analysis covering implant length, cortical thickness, bone loss and transducer fixation [14], a Kriging surrogate model of resonance frequency built from Latin hypercube finite element runs [24], a model that converts simulated micromobility to ISQ across bone grades [25], and a recent analysis of ISQ and bone stress for tilted implants [26]. In vitro studies in resin or polyurethane varied cortical thickness and trabecular properties [13, 27]. A review analyzed the physical models behind clinical stability devices and discussed what each actually measures [28]. An analytical model of the ASIST system treats the implant as a vibrating linear system and reports that fitting two vibration modes gives an interfacial stiffness that does not change with abutment type in foam [29]. Damping and resonance frequency have also been related mathematically [30].

Consequently, the proposition “ISQ depends on cortical stiffness and exposed height” is established, and we do not claim it. What we could not find in the searched sources was: (a) a one-page lumped model with closed-form sensitivities intended for reasoning about placement thresholds; (b) an explicit treatment of how such construct differences bear on prognostic validity, including the conditions under which they would not; and (c) a smallest-detectable-change framework applied to the dip that distinguishes individual from group-level detectability. The search covered PubMed and Europe PMC only; Embase, Scopus and Web of Science were inaccessible, so absence of a hit is weak evidence. Patents listed in the underlying notes (US5392779A and earlier) could not be verified and are not relied upon.

Prognostic validity of the thresholds

The earliest support found for a torque rule was a small series of single-tooth implants in which most immediately loaded failures occurred among implants placed at 20 N·cm [31]; how 35 N·cm entered guidelines was not established here. Diagnostic accuracy of RFA for failure of immediately loaded implants was low, with pooled sensitivity 0.38, specificity 0.73 and AUC 0.54 [5]. A meta-analysis of low insertion torque in immediately loaded implants found 96% survival at or below 35 N·cm and no significant effect [6]. A meta-analysis of non-immediately loaded implants found survival of 97.7% with ISQ below 60 and 95.2% with torque below 35 N·cm, and no threshold effect [7]. A large prospective series of 4,114 implants found that clinically unstable implants survived at 97.2% and that placement ISQ was unrelated to survival [32]. Low-torque implants also integrate in a closed cohort [33].

The opposing evidence is limited but real. One retrospective series found that all failures lay below ISQ 66 to 67 [34]. A small cohort found that ISQ at 8 weeks, not at placement, discriminated failures with an AUC of 0.93 [16]. The latter is compatible with our hypothesis, because by 8 weeks the system reflects healed interface, and the former is compatible with a weak, threshold-like safety signal at low values that mean AUC would understate. Device agreement is a further issue: a damping-based device and a resonance device correlated only at about 0.4 in 890 implants and misclassified in the low range [35].

The stability dip

In a prospective series of 27 implants, stability was lowest at 3 weeks, with the largest decrease (8.6%) in type 4 bone [36]. In 25 implants measured with an RFA device, a 3 to 4 unit decline at 3 weeks showed no surface effect [8], whereas in a randomized pilot of 31 patients the turning point occurred at 2 versus 4 weeks depending on surface [37]. A trial in the posterior maxilla found a dip with conventional drilling and none with osseodensification [38]. In sheep, RFA did not detect early biomechanical changes that histology showed [39]. Conversely, direct displacement sensing in rabbits showed a real early loss of stability under immediate loading that ISQ change tracked closely [40]. The dip is therefore inconsistent in timing, size and presence, and its direct biological verification in humans is lacking.

Reliability data are mixed. In one clinical study ICC was 0.97 for repeated Osstell readings, yet readings were identical or within three ISQ points in only 52.9% and 62.4% of cases for two transducers [41]. An in vitro study found repeatability of 0.90 for one device and 0.60 for another, with reliable readings only in stiff embedding materials [42]. Inter-device agreement was 82% within 4 units for one pair of instruments, although the mean difference was judged clinically unimportant by the author [18].

3. Hypothesis and theoretical framework

Formal statement

H1 (construct hypothesis). For a given transducer and implant, the placement-time resonance frequency f is a monotone increasing function of the tip stiffness k_(tip) of the system, and k_(tip) is dominated by two quantities: the free length L of the vibrating system above the supporting bone and the rotational stiffness k_θ at the crest, which is itself dominated by the crestal cortex. Trabecular stiffness contributes to k_θ but with a smaller weight than the cortex over a realistic range.

H2 (decoupling hypothesis). Insertion torque T is a function of frictional and compressive resistance integrated along the whole embedded length, with weights on cortex and trabecular bone different from those in k_θ. Hence corr(f,T) is positive but moderate, and the residual variance of each contains information the other lacks.

H3 (prognostic hypothesis). If late failure depends mainly on a latent interface property q (for example trabecular-contact quality or bone vitality) that enters k_θ only partly, then placement ISQ discriminates failure poorly, and discriminates it less well than a variable weighted more heavily toward q. If failure depends instead on total k_θ, ISQ would discriminate acceptably, and the hypothesis is refuted for that mechanism.

H4 (dip hypothesis). The reported 2 to 4 week decline of 3 to 4 ISQ units is comparable to the smallest detectable change (SDC) of an individual reading for plausible reliability. It may nonetheless be detectable as a mean change in a sufficiently large paired sample, provided that systematic sources of bias (peg orientation, device, abutment height) are controlled.

Assumptions

The model assumes: (A1) the first bending mode dominates the reading, with the transducer rigidly attached to the implant; (A2) the implant and transducer behave as a uniform Euler-Bernoulli beam of flexural rigidity EI; (A3) the bone provides a single lumped rotational spring at the crest, the sum of a cortical part and a trabecular part; (A4) translational compliance, interface damping and higher modes are ignored; (A5) the properties are linear and constant during the measurement; (A6) mass is lumped in the free length. Assumption A4 is the most restrictive, because the ASIST work shows that second-mode information is informative about interface stiffness independent of components [29]. A1 is an approximation for devices that excite the transducer magnetically and read the first resonance.

Equations

Let L (m) be the effective free length, E (Pa) Young’s modulus of the implant-transducer material, I=πd^4/64 (m^4) the second moment of area for an equivalent solid cylinder of diameter d (m), k_θ (N·m/rad) the rotational stiffness of the bone support, and m (kg) the mass of the free portion. The tip stiffness of a cantilever on a rotational spring is the series combination of beam bending and rotation:

1/(k_(tip))=(L^3)/3EI+(L^2)/k_θ

The resonance frequency follows from an effective mass m_(eff), for which a uniform distributed-mass cantilever gives the Rayleigh factor 33/140:

f=1/2π√((k_(tip))/(m_(eff))),  m_(eff)=33/140 m

The rotational stiffness is split into cortical and trabecular parts, each scaled to baseline by a stiffness-times-extent product:

k_θ=k_c+k_t,  k_c=(1−s) K_0 E_ct_c/(E_(c0)t_(c0)),  k_t=s K_0 E_tl_b/(E_(t0)l_(b0))

where E_c and t_c are cortical modulus and thickness, E_t and l_b trabecular modulus and embedded length, s is the trabecular share of k_θ at baseline (an assumption), and K_0 is the baseline rotational stiffness, here calibrated so that the baseline frequency equals 7,000 Hz. The linear scaling in modulus times extent simplifies the true contact mechanics.

Two consequences are independent of the numbers. First, because k_(tip) falls with L^(−2) at least, frequency is highly sensitive to L: any change in exposed height, including marginal bone loss or a taller abutment, moves frequency irrespective of interface health. Second, since k_θ enters through a sum, a stiff crestal cortex can compensate for a soft trabecular bed and vice versa, so the same frequency can arise from different interfaces. This is the central point behind decoupling from torque, which is more sensitive to the embedded trabecular zone.

Illustrative computation under stated assumptions — not empirical data

The script rfa_model.py (saved with this manuscript, run on 17 November 2026) evaluates the model. Assumed parameters: titanium E=110 GPa; equivalent diameter d=4.1 mm, giving I=13.87 mm^4 and EI=1.526 N·m²; free mass m=2.0 g, giving m_(eff)=0.471 g; baseline free length L_0=10 mm; baseline cortical modulus 13.7 GPa and thickness 1.5 mm; trabecular modulus 0.5 GPa and embedded length 10 mm. These values are chosen for plausibility, not measured. Calibration to 7,000 Hz gave k_(tip,0)=9.12×10^5 N/m and K_0=113.9 N·m/rad, with beam bending contributing 19.9% of tip compliance and the rotational spring the rest. That split follows from the assumed m, E and d, not from data. Only relative sensitivities are interpretable, and they depend on the assumed split.

Table 2. Local elasticities ∂lnf/∂lnx at baseline (assumed parameters; not empirical data).
Quantitys = 0.15s = 0.30s = 0.50
Free length L-1.100-1.100-1.100
Cortical modulus or thickness+0.340+0.280+0.200
Trabecular modulus or embedded length+0.060+0.120+0.200

One-at-a-time scenarios at s=0.30 (baseline 7,000 Hz) gave: free length +1 mm, 6,301 Hz (-699 Hz, -10.0%); -1 mm, 7,856 Hz (+12.2%); +2 mm, 5,720 Hz (-18.3%); cortical thickness 1.0 mm, 6,277 Hz (-10.3%); 2.5 mm, 8,109 Hz (+15.8%); no cortex, 4,133 Hz (-41.0%); trabecular modulus x0.5, 6,552 Hz (-6.4%); x2, 7,753 Hz (+10.8%); x0.2, 6,254 Hz (-10.7%).

The output shows that, under these assumptions, a 1 mm change in free length shifts frequency about as much as a fivefold fall in trabecular modulus, and about as much as a 1 mm change in cortical thickness. The elasticity of free length (-1.1) exceeds that of any bone property; the elasticity of the trabecular terms reaches the cortical value only when the trabecular share is set to 0.50. Mapping a frequency change onto ISQ units requires the proprietary conversion, which we did not have; all conclusions about ISQ units below therefore rely on published ISQ-scale reliability, not on this model.

Monte Carlo coupling of the model to torque and failure

To see how construct mismatch could affect discrimination, 200,000 virtual implants were drawn with assumed independent lognormal variation: cortical modulus (coefficient of variation 0.2), cortical thickness (0.4), trabecular modulus (0.5); free length normal with SD 0.8 mm (clipped to 7 to 14 mm); and relative frequency noise of 1% SD. A torque proxy was defined as 35 N·cm times a weighted sum of the cortical (weight 0.35) and trabecular (weight 0.65) stiffness ratios with lognormal noise (CV 0.15). Failure probability followed a logistic function of the standardized latent variable, with intercept -3.2 and slope -1.0 (failure rate about 5.8%). Two scenarios for the latent driver were run: trabecular modulus alone, or total k_θ. These weights and slopes are assumptions chosen so that torque tracks the trabecular bed more than frequency does, which builds in part of the hypothesis. The simulation shows consistency, not support.

Table 3. Monte Carlo output (assumed parameters; not empirical data). AUC is the area under the ROC curve for the placement variable against failure.
Trabecular share sFailure driverr(f, T)AUC of fAUC of TAUC of the latent driver
0.15Trabecular modulus0.470.5470.6990.754
0.15Total k_theta0.470.7180.6520.718
0.30Trabecular modulus0.590.5870.6980.758
0.30Total k_theta0.590.7140.6930.714
0.50Trabecular modulus0.710.6440.6990.756
0.50Total k_theta0.700.7110.7280.711

The simulated correlation between frequency and torque (0.47 to 0.71) lies at or above the pooled empirical value of 0.44 reported above, so the model produces a moderate correlation without being fitted to it, though not as low as the pooled figure. When failure is driven by the trabecular bed, frequency discriminates poorly (AUC 0.55 to 0.64) and torque better (about 0.70), and both fall short of the latent driver itself (about 0.75) because of noise and mixing. When failure is driven by total k_θ, frequency reaches its ceiling (0.71) and the advantage of torque disappears. Thus the model does not by itself predict that thresholds fail. It predicts that the validity of ISQ depends on which latent quantity governs failure, and that a low empirical AUC is expected when that quantity is not the crestal bending stiffness. The empirical AUC of 0.54 [5] is closer to the first scenario, but a low failure rate, heterogeneous populations and outcome definitions could also produce it.

Measurement error and the dip

Let σ_B be the between-implant SD of ISQ in a cohort and ICC the reliability. The standard error of measurement is SEM=σ_B√(1−ICC) and the smallest detectable change for an individual at 95% confidence is SDC=1.96√(2) SEM. Between-implant SDs of 6.23 and 9.43 ISQ units have been reported for secondary and primary stability in one clinical series [20]; we used these as assumed σ_B values. ICC values were taken from published reliability: 0.97 for repeated readings in one clinical study [41], 0.90 and 0.60 for two devices in one in vitro study [42].

Computed SDC values (ISQ units; assumed σ_B and published ICC; not empirical data) were: σ_B 6.2 with ICC 0.97, SEM 1.07 and SDC 2.98; σ_B 9.4 with ICC 0.97, SEM 1.63 and SDC 4.51; σ_B 6.2 with ICC 0.90, SEM 1.96 and SDC 5.43; σ_B 9.4 with ICC 0.90, SEM 2.97 and SDC 8.24; and σ_B 9.4 with ICC 0.60, SEM 5.95 and SDC 16.48.

For a reported dip of 3 to 4 units [8], the individual SDC exceeds the dip for every combination except the most favorable (ICC 0.97 with σ_B=6.2, SDC 2.98). Under SEM values consistent with ICC 0.97 and ICC 0.90, the proportion of implants showing an apparent decrease of at least 3 units from noise alone would be 9.2% when SEM is 1.6 and 24.0% when SEM is 3.0; for a drop of at least 4 units, 3.9% and 17.3%. The reliability figures of Herrero-Climent [41] illustrate the same point from the other direction: ICC 0.97 coexists with readings that differ by three points or more in roughly 38 to 47% of cases (100 minus the reported 62.4% and 52.9%). However, a group mean can be informative. For a paired design with 80% power and two-sided alpha 0.05, a true mean dip of 3.0 units requires 9 implants when SEM is 1.6 and the true between-implant SD of change is 2.0, and 25 when SEM is 3.0 and that SD is 3.0. For a dip of 4.0 units the corresponding sizes are 6 and 15. Group-level detectability is therefore plausible; the existing small series (20 to 31 implants) are of the right size but are exposed to systematic bias, and the dip is not interpretable in an individual patient.

Numbered predictions and refutation criteria

P1. In a factorial bench experiment with fixed trabecular foam, a 1 mm increase in exposed length lowers frequency by an amount comparable to a 1 mm decrease in cortical shell thickness, with the length elasticity near -1 to -1.5 on the log-log scale. Refuted if the length elasticity magnitude is below 0.5 or the cortex elasticity exceeds that of length by a factor of 2 or more.

P2. The cortical-shell effect on frequency is larger than the trabecular-foam effect over the physiological range of each, with the ratio depending on the cortex to foam stiffness ratio. Refuted if foam modulus explains more frequency variance than the cortical shell in a design spanning clinically realistic ranges for both.

P3. In cadaver or ex vivo human bone with micro-CT, variance partitioning assigns most ISQ variance to exposed height and crestal cortical thickness, and a smaller share to trabecular BV/TV and contact fraction. Refuted if BV/TV or contact fraction explains independent variance in ISQ greater than that of cortex after adjusting for height.

P4. Torque is predicted better by embedded-length-integrated variables (trabecular BV/TV, thread engagement) than ISQ is, in the same specimens. Refuted if the incremental explained variance of BV/TV for torque is not larger than for ISQ.

P5. Within one site, ISQ measured on the abutment, on the platform and with a taller peg shifts in the direction and approximate proportion predicted by the model with L. Refuted if the shifts are unrelated to L after control for device.

P6. In pooled individual-participant data, placement ISQ and torque each reach AUC below 0.70 for failure within each loading protocol, and the correlation r(ISQ,T) lies between 0.3 and 0.6. Refuted if either threshold exceeds AUC 0.70 with stable calibration across devices.

P7. Adding exposed height and crestal cortical thickness to a model of ISQ removes most of the ISQ-torque correlation (partial correlation falls substantially). Refuted if the partial correlation remains at or near the raw value.

P8. In repeated-measures clinical data with triplicate readings, the individual SDC for ISQ is at least as large as the mean 2 to 4 week change in most cohorts, while the paired group mean change remains detectable. Refuted if the SDC is clearly smaller than the reported dip in most cohorts, or if the dip is absent in a phantom-adjusted analysis.

4. Proposed methodology

The program has four phases. Each phase targets specific predictions, and none has been performed.

Phase 1: factorial bench experiment (P1, P2, P5)

Design. A full factorial of exposed height (at least three levels, achieved by pegs or abutments of known length and by cutting the crestal offset), cortical shell (at least three thicknesses including zero, in a synthetic cortical material of known modulus), and cancellous-simulating foam (at least three densities with known modulus), with replicate implants of a single design and a second design for generalization. Specimens are held in a rigid fixture with a defined boundary to avoid fixture resonance. Hsu and colleagues used four cortical thicknesses and five foam strengths in 60 specimens with torque, ISQ and Periotest [13], and Feng and colleagues varied cortical height and trabecular status in epoxy resins [27]. The present design adds exposed length as an independent factor and recovers the frequency itself where the device allows.

Measurements. Raw resonance frequency if the instrument exposes it; ISQ otherwise. Two commercial devices, three readings per orientation, two orthogonal orientations, with the transducer re-seated between readings. Insertion torque recorded as a full curve with a calibrated driver at constant speed. Because the ISQ map is proprietary, fitted elasticities are computed on ISQ and on frequency when available; if only ISQ is available, the model comparison is made on the monotone-transformed scale and the conclusions are limited to rank-order and sign.

Analysis. Log-log regression of frequency (or ISQ) on length, cortical stiffness-thickness product and foam modulus, with interactions; nonlinear least squares fit of the lumped model with K_0, s and the beam-to-support split as free parameters; comparison with a three-dimensional finite element model and with the second-mode analytical model of the ASIST type [29]. The lumped model is accepted as a useful summary only if its out-of-sample prediction error (leave-one-factor-level-out) is within the replicate measurement error.

Sample size. No inferential sample size is set because the quantity of interest is a regression elasticity with high bench signal-to-noise; a planning heuristic (not derived) of 10 specimens per fitted parameter implies 50 to 70 specimens, to be replaced by a simulation-based precision calculation after pilot work.

Phase 2: cadaver variance partitioning (P3, P4, P7)

Design. Human cadaver jaws, such as those used in earlier cadaver work [4, 11], with implants placed by a standardized protocol at multiple sites per jaw, using a recorded torque curve. Micro-CT or CBCT before and after placement yields crestal cortical thickness, trabecular BV/TV, thread-bone contact fraction and exposed height (peg length plus crestal offset). Readings are taken with two devices at fixed orientations.

Analysis. Mixed models with specimen as a random effect and dominance analysis to apportion explained variance of ISQ and torque to the predictor groups: exposed height, cortical, trabecular. The key contrast is the difference between the cortical and trabecular dominance statistics for ISQ versus for torque (P3, P4). Partial correlations of ISQ and torque after adjustment for height and cortex test P7. Because ex vivo bone lacks perfusion and soft-tissue constraints, results will be limited to mechanical construct questions.

Phase 3: repeated-measures reliability and dip study (P5, P8)

Design. A prospective observational study at one or more centers. Participants receiving implants have ISQ measured in triplicate with two devices at placement and at weekly intervals through week 8, with a standardized peg and orientation. A non-implant phantom (an implant in a stiff block, such as resin) is measured at every session to capture device drift and operator effects. Abutment or peg height, occlusal contact, antagonist status and loading protocol are recorded because abutment height alone changed ISQ substantially in one clinical comparison [15] and may interact with occlusion [22]. Implants that are loaded early are analyzed separately because loading alters stability [40].

Estimands. (i) Within-implant repeatability SD and SEM by device. (ii) Between-device limits of agreement. (iii) SDC for individuals. (iv) The mean change from baseline to the nadir week and the proportion of implants whose change exceeds the SDC.

Analysis. Variance components from a mixed model with implant, session and device random effects give ICC and SEM. Bland-Altman analysis with limits of agreement for device comparisons, following the agreement framing used in earlier instrument comparisons [18]. Mean change by week with a mixed model including phantom reading as covariate to adjust drift.

Sample size. For the paired mean dip, the formula used above is n=((z_(1−α/2)+z_(1−β)) σ_Δ/δ)^2+1, with σ_Δ=√(2 SEM^2+σ_h^2) where σ_h is the true between-implant SD of change and δ the dip. For α=0.05, power 0.80, δ=3 to 4 units, SEM 1.6 to 3.0 and σ_h=2 to 3, the computation above gave 6 to 25 implants. Because loss to follow-up, dropout by loading, and clustering by patient are expected, and because reliability estimates themselves need more observations for precision, we propose at least 60 implants as a working target; the SEM must be re-estimated from pilot data before this is fixed.

Phase 4: individual-participant reanalysis of thresholds (P6)

Design. A meta-research collaboration to obtain individual-participant data from placement-time torque and ISQ cohorts with survival or failure follow-up, preserving device identity and loading protocol. Where data cannot be pooled, a registry or institutional cohort may serve.

Analysis. ROC curves with confidence intervals by loading protocol and device, calibration plots, and decision-curve analysis to estimate the net benefit and misclassification cost of each threshold, since discrimination alone does not answer whether a threshold changes decisions. A model with torque, ISQ, exposed height and cortical thickness (when CBCT exists) tests whether adding geometry improves discrimination. Failure events are rare in many cohorts (survival around 95 to 98% in the meta-analyses cited above [6, 7]), so the precision of AUC depends on the number of failures, not implants; with a 3% failure rate, 1,000 implants yield about 30 events, which is a modest number for a stable AUC estimate.

Bias control, ethics and feasibility

Operators and analysts are blinded to the model predictions and, in Phase 3, readings are recorded before knowledge of previous values. Device calibration is verified with a standard stiff block at every session. Phase 2 requires institutional approval for cadaver material. Phase 3 requires ethics approval and informed consent; none has been obtained or is claimed. Phase 4 requires data-sharing agreements.

5. Competing explanations and how the design discriminates among them

Four competing explanations were considered.

Alternative 1: ISQ broadly reflects interface stiffness, and weak validity is noise. Under this view, trabecular stiffness and contact fraction should explain ISQ variance independently of cortex and height. Phase 2 (P3) discriminates directly, and Phase 1 (P2) provides a mechanical test. The synthetic-bone evidence already shows trabecular sensitivity [13], so a result in which trabecular terms carry substantial weight would partly support this alternative and reduce the contrast between ISQ and torque. The hypothesis is then not refuted entirely but reduced to a quantitative claim about relative weight.

Alternative 2: weak prognostic validity reflects rare failure and outcome heterogeneity, not mechanics. Low event counts depress the precision and bias the AUC; range restriction (clinicians avoid immediate loading in low-ISQ sites) truncates the predictor. This explanation is not mutually exclusive. Phase 4 can discriminate partially, because it can examine whether adding geometry improves discrimination. Range restriction cannot be fully removed from observational data, and this is stated as a limit.

Alternative 3: the dip is a real biological event, and ISQ tracks it. Direct displacement sensing supports a real early loss in loaded rabbit implants with close ISQ tracking [40]. The sheep study found the opposite for unloaded implants [39]. Phase 3 does not deny a biological dip; it asks whether the dip can be separated from measurement and systematic bias in individuals. A phantom-adjusted group mean change larger than the SDC-derived noise would support a real dip at the group level, a result compatible with the hypothesis as stated in H4.

Alternative 4: low torque-ISQ correlation reflects driving-protocol and device artifacts. Hsu’s synthetic bone results show both measures respond to the same two material variables in controlled conditions [13]. Phase 1 uses a controlled driver and two devices, and Phase 2 records torque curves, so that artifact variance can be separated from construct variance in the specimens. In the clinic, drilling protocols differ widely (an osseodensification trial reported different dip behavior by technique [38]); this cannot be eliminated and is recorded as a covariate.

6. Clinical and engineering implications

The following are conditional on the hypotheses surviving the validation in Section 4.

For clinicians. If exposed height and crestal cortex dominate placement ISQ, then an ISQ value at placement should be read with the recorded transducer height, abutment or peg type, and site (mandible versus maxilla) in mind. The abutment-versus-platform data show that readings at different heights cannot be compared [15]. A single ISQ cut-off applied across heights and devices would then merge different mechanical conditions. Comparison of serial readings within one implant, using the same transducer and orientation, would be more interpretable than cross-patient comparison with a fixed threshold, but a within-implant change below the individual SDC should not be called a dip.

For loading decisions. Existing meta-analyses do not support fixed thresholds as stand-alone survival predictors [5, 6, 7, 32]. If thresholds are retained as safety heuristics, their use as a low-value alarm (a very low reading prompting review) is not contradicted by the data in this paper, because the model predicts weaker discrimination across the whole range and does not exclude a signal at the extremes. The retrospective finding that all failures lay below ISQ 66 to 67 [34] is compatible with such an alarm role.

For manufacturers and engineers. Reporting the underlying frequency, and the mapping to ISQ, would let investigators separate geometry from stiffness. Multi-mode analysis, as in the analytical ASIST work, offers a way to estimate interface stiffness independent of the abutment [29]. A height-corrected stability index, or a standardized reference peg, would remove one of the largest systematic sources of variation identified in the model. Weak agreement between damping and frequency devices in the low range [35] supports reporting device identity with every value.

7. Limitations

Model structure. The lumped model ignores translational compliance, interface damping, the three-dimensional shape of the crestal defect, bucco-lingual versus mesio-distal mode differences, and higher modes. Bucco-lingual readings were less reliable than other directions in a three-instrument comparison [18], which a one-dimensional bending model cannot represent. The multi-mode analytical approach shows that a single-frequency reading conflates interface and component properties [29], and our single-frequency model inherits that conflation.

Parameter choice. The beam share of compliance (19.9%), the trabecular share s, and the torque weights were assumed. The elasticities in Table 2 move with s (cortical elasticity 0.34 to 0.20 as s rises from 0.15 to 0.50), and the Monte Carlo AUC values depend on the assumed failure driver and weights. The simulation embeds the claim that torque weights the trabecular zone more than frequency does; it therefore cannot support that claim. No parameter was fitted to data, and agreement of the simulated ISQ-torque correlation with published values is partial (the simulated values of 0.47 to 0.71 are at or above the pooled 0.44 [1]).

ISQ scale. We did not have the manufacturer’s frequency-to-ISQ mapping, so translating predicted frequency shifts into ISQ units was not done. The SDC calculations use published ISQ-scale reliability and are independent of the model. The σ_B values (6.23 and 9.43) come from one clinical series [20] and may not transfer.

Evidence base. Appraisal was at abstract level. Reliability statistics in the cited studies (ICC and repeatability) were taken from abstracts that do not specify all design details. The ICC is sample-dependent and its translation to SEM requires the between-implant SD of the same sample, which we imported from a different study; the resulting SDC values are order-of-magnitude illustrations. One paper on implant diameter and length in polyurethane blocks was flagged as retracted in PubMed and was excluded.

Counter-evidence. Direct displacement sensing [40] and healed-interface micromotion in sheep [17] suggest that ISQ tracks real mechanical change once the interface has matured. The hypothesis applies chiefly to placement-time readings and does not claim that RFA is uninformative in healing.

8. Conclusions

The available evidence supports a construct explanation for why placement ISQ and insertion torque correlate only moderately: resonance frequency is strongly sensitive to the unsupported length of the implant system and to the crestal cortex, whereas torque integrates resistance over the embedded length. This mechanical dependence is already established in the finite element, bench and clinical literature, so the contribution of this manuscript is the explicit linkage to prognostic validity and to measurement error. Under the stated assumptions, a lumped model reproduces moderate ISQ-torque correlation and shows that poor discrimination of failure by ISQ is expected only if failure depends on an interface property that frequency weights weakly. It is not expected if failure tracks total crestal stiffness. The 3 to 4 unit dip is of the same size as the individual smallest detectable change under plausible reliability and may still be detectable as a group mean change. These statements are hypotheses supported by calculation under assumptions, not by data, and they are open to refutation by the eight predictions in Section 3.6.

Cite this paper

Jain S, Gupta A. What Does the Implant Stability Quotient Measure? A Mechanical Model of Resonance Frequency Analysis and Its Consequences for Loading Thresholds. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/isq-mechanical-model

References

  1. Tisci A, Fanelli F, Caponio VCA, Zhurakivska K, Dioguardi M, Troiano G. Resonance Frequency Analysis and Clinical Outcomes in Implant Dentistry: A Systematic Review and Meta-Analysis. Clin Implant Dent Relat Res. 2026;28(3):e70156. doi:10.1111/cid.70156 (opens in a new tab). PMID: 42117690 (opens in a new tab).
  2. Lages FS, Douglas-de Oliveira DW, Costa FO. Relationship between implant stability measurements obtained by insertion torque and resonance frequency analysis: A systematic review. Clin Implant Dent Relat Res. 2018;20(1):26-33. doi:10.1111/cid.12565 (opens in a new tab). PMID: 29194944 (opens in a new tab).
  3. Manresa C, Bosch M, Echeverría JJ. The comparison between implant stability quotient and bone-implant contact revisited: an experiment in Beagle dog. Clin Oral Implants Res. 2014;25(11):1213-1221. doi:10.1111/clr.12256 (opens in a new tab). PMID: 24102812 (opens in a new tab).
  4. Rozé J, Babu S, Saffarzadeh A, Gayet-Delacroix M, Hoornaert A, Layrolle P. Correlating implant stability to bone structure. Clin Oral Implants Res. 2009;20(10):1140-5. doi:10.1111/j.1600-0501.2009.01745.x (opens in a new tab). PMID: 19519789 (opens in a new tab).
  5. Atieh MA, Alsabeeha NH, Payne AG. Can resonance frequency analysis predict failure risk of immediately loaded implants?. Int J Prosthodont. 2012;25(4):326-39. PMID: 22720282 (opens in a new tab).
  6. Darriba I, Seidel A, Moreno F, Botelho J, Machado V, Mendes JJ, et al. Influence of low insertion torque values on survival rate of immediately loaded dental implants: A systematic review and meta-analysis. J Clin Periodontol. 2023;50(2):158-169. doi:10.1111/jcpe.13733 (opens in a new tab). PMID: 36217696 (opens in a new tab).
  7. Razaghi N, Gill T, Patankar A, Petrie A, Shahdad S. Significance of Primary Stability in Nonimmediately Loaded Dental Implants: A Systematic Review and Meta-analysis. Int J Prosthodont. 2025;38(4):457-482. doi:10.11607/ijp.9115 (opens in a new tab). PMID: 40172942 (opens in a new tab).
  8. Han J, Lulic M, Lang NP. Factors influencing resonance frequency analysis assessed by Osstell mentor during implant tissue integration: II. Implant surface modifications and implant diameter. Clin Oral Implants Res. 2010;21(6):605-11. doi:10.1111/j.1600-0501.2009.01909.x (opens in a new tab). PMID: 20666787 (opens in a new tab).
  9. Meredith N, Alleyne D, Cawley P. Quantitative determination of the stability of the implant-tissue interface using resonance frequency analysis. Clin Oral Implants Res. 1996;7(3):261-7. doi:10.1034/j.1600-0501.1996.070308.x (opens in a new tab). PMID: 9151590 (opens in a new tab).
  10. Meredith N, Shagaldi F, Alleyne D, Sennerby L, Cawley P. The application of resonance frequency measurements to study the stability of titanium implants during healing in the rabbit tibia. Clin Oral Implants Res. 1997;8(3):234-43. doi:10.1034/j.1600-0501.1997.080310.x (opens in a new tab). PMID: 9586468 (opens in a new tab).
  11. O’Sullivan D, Sennerby L, Meredith N. Measurements comparing the initial stability of five designs of dental implants: a human cadaver study. Clin Implant Dent Relat Res. 2000;2(2):85-92. doi:10.1111/j.1708-8208.2000.tb00110.x (opens in a new tab). PMID: 11359268 (opens in a new tab).
  12. Sennerby L, Meredith N. Implant stability measurements using resonance frequency analysis: biological and biomechanical aspects and clinical implications. Periodontol 2000. 2008;47:51-66. doi:10.1111/j.1600-0757.2008.00267.x (opens in a new tab). PMID: 18412573 (opens in a new tab).
  13. Hsu JT, Fuh LJ, Tu MG, Li YF, Chen KT, Huang HL. The effects of cortical bone thickness and trabecular bone strength on noninvasive measures of the implant primary stability using synthetic bone models. Clin Implant Dent Relat Res. 2013;15(2):251-61. doi:10.1111/j.1708-8208.2011.00349.x (opens in a new tab). PMID: 21599830 (opens in a new tab).
  14. Winter W, Möhrle S, Holst S, Karl M. Parameters of implant stability measurements based on resonance frequency and damping capacity: a comparative finite element analysis. Int J Oral Maxillofac Implants. 2010;25(3):532-9. PMID: 20556252 (opens in a new tab).
  15. Lages FS, Willya Douglas-de-Oliveira D, Ibelli GS, Assaf F, Queiroz TP, Costa FO. Relationship between implant stability on the abutment and platform level by means of resonance frequency analysis: A cross-sectional study. PLoS One. 2017;12(7):e0181873. doi:10.1371/journal.pone.0181873 (opens in a new tab). PMID: 28742164 (opens in a new tab).
  16. Atieh MA, Alsabeeha NH, Payne AG, de Silva RK, Schwass DS, Duncan WJ. The prognostic accuracy of resonance frequency analysis in predicting failure risk of immediately restored implants. Clin Oral Implants Res. 2014;25(1):29-35. doi:10.1111/clr.12057 (opens in a new tab). PMID: 23113597 (opens in a new tab).
  17. Trisi P, Berardini M, Falco A, Podaliri Vulpiani M. Validation of value of actual micromotion as a direct measure of implant micromobility after healing (secondary implant stability). An in vivo histologic and biomechanical study. Clin Oral Implants Res. 2016;27(11):1423-1430. doi:10.1111/clr.12756 (opens in a new tab). PMID: 26727557 (opens in a new tab).
  18. Norton MR. Resonance Frequency Analysis: Agreement and Correlation of Implant Stability Quotients Between Three Commercially Available Instruments. Int J Oral Maxillofac Implants. 2019;34(1):215–222. doi:10.11607/jomi.6964 (opens in a new tab). PMID: 30282089 (opens in a new tab).
  19. Chatvaratthana K, Thaworanunta S, Seriwatanachai D, Wongsirichat N. Correlation between the thickness of the crestal and buccolingual cortical bone at varying depths and implant stability quotients. PLoS One. 2017;12(12):e0190293. doi:10.1371/journal.pone.0190293 (opens in a new tab). PMID: 29281715 (opens in a new tab).
  20. Tanaka K, Sailer I, Iwama R, Yamauchi K, Nogami S, Yoda N, et al. Relationship between cortical bone thickness and implant stability at the time of surgery and secondary stability after osseointegration measured using resonance frequency analysis. J Periodontal Implant Sci. 2018;48(6):360-372. doi:10.5051/jpis.2018.48.6.360 (opens in a new tab). PMID: 30619637 (opens in a new tab).
  21. Miyamoto I, Tsuboi Y, Wada E, Suwa H, Iizuka T. Influence of cortical bone thickness and implant length on implant stability at the time of surgery–clinical, prospective, biomechanical, and imaging study. Bone. 2005;37(6):776-80. doi:10.1016/j.bone.2005.06.019 (opens in a new tab). PMID: 16154396 (opens in a new tab).
  22. Martín-Martín A, Pérez-Pevida E, Férnandez-Hernández S, Lubillo-Valdeón J, Brizuela-Velasco A. Influence of Healing Abutment Height on Secondary Implant Stability Using Resonance Frequency Analysis: A Prospective Clinical Study. J Clin Med. 2025;14(14). doi:10.3390/jcm14145140 (opens in a new tab). PMID: 40725834 (opens in a new tab).
  23. Monje A, Ravidà A, Wang HL, Helms JA, Brunski JB. Relationship Between Primary/Mechanical and Secondary/Biological Implant Stability. Int J Oral Maxillofac Implants. 2019;34:s7-s23. doi:10.11607/jomi.19suppl.g1 (opens in a new tab). PMID: 31116830 (opens in a new tab).
  24. Chu L, Shi J, Souza de Cursi E. Kriging Surrogate Model for Resonance Frequency Analysis of Dental Implants by a Latin Hypercube-Based Finite Element Method. Appl Bionics Biomech. 2019;2019:3768695. doi:10.1155/2019/3768695 (opens in a new tab). PMID: 31093299 (opens in a new tab).
  25. Pammer D. Evaluation of postoperative dental implant primary stability using 3D finite element analysis. Comput Methods Biomech Biomed Engin. 2019;22(3):280-287. doi:10.1080/10255842.2018.1552682 (opens in a new tab). PMID: 30676075 (opens in a new tab).
  26. Ceddia M, Romasco T, Marchioli G, Comuzzi L, Cipollina A, Piattelli A, et al. Finite Element Analysis of Implant Stability Quotient (ISQ) and Bone Stresses for Implant Inclinations of 0°, 15°, and 20°. Materials (Basel). 2025;18(7). doi:10.3390/ma18071625 (opens in a new tab). PMID: 40271905 (opens in a new tab).
  27. Feng SW, Lin CT, Chang WJ, Lee SY, Cheng CH, Huang HM. Influence of simulated bone quality and cortical bone thickness on implant stability detection using resonance frequency and damping factor analysis. Int J Oral Maxillofac Implants. 2014;29(1):105-12. doi:10.11607/jomi.3181 (opens in a new tab). PMID: 24451860 (opens in a new tab).
  28. Zanetti EM, Pascoletti G, Calì M, Bignardi C, Franceschini G. Clinical Assessment of Dental Implant Stability During Follow-Up: What Is Actually Measured, and Perspectives. Biosensors (Basel). 2018;8(3). doi:10.3390/bios8030068 (opens in a new tab). PMID: 30011870 (opens in a new tab).
  29. Jar C, Archibald A, Gibson M, Westover L. An analytical model to measure dental implant stability with the Advanced System for Implant Stability Testing (ASIST). J Mech Behav Biomed Mater. 2024;150:106238. doi:10.1016/j.jmbbm.2023.106238 (opens in a new tab). PMID: 37992584 (opens in a new tab).
  30. Kwon TK, Kim HY, Yang JH, Wikesjö UM, Lee J, Koo KT, et al. First-Order Mathematical Correlation Between Damping and Resonance Frequency Evaluating the Bone-Implant Interface. Int J Oral Maxillofac Implants. 2016;31(5):1008-15. doi:10.11607/jomi.4468 (opens in a new tab). PMID: 27632254 (opens in a new tab).
  31. Ottoni JM, Oliveira ZF, Mansini R, Cabral AM. Correlation between placement torque and survival of single-tooth implants. Int J Oral Maxillofac Implants. 2005;20(5):769-76. PMID: 16274152 (opens in a new tab).
  32. Rodrigo D, Aracil L, Martin C, Sanz M. Diagnosis of implant stability and its impact on implant survival: a prospective case series study. Clin Oral Implants Res. 2010;21(3):255-61. doi:10.1111/j.1600-0501.2009.01820.x (opens in a new tab). PMID: 19958375 (opens in a new tab).
  33. Norton MR. The Influence of Low Insertion Torque on Primary Stability, Implant Survival, and Maintenance of Marginal Bone Levels: A Closed-Cohort Prospective Study. Int J Oral Maxillofac Implants. 2017;32(4):849-857. doi:10.11607/jomi.5889 (opens in a new tab). PMID: 28708918 (opens in a new tab).
  34. Baltayan S, Pi-Anfruns J, Aghaloo T, Moy PK. The Predictive Value of Resonance Frequency Analysis Measurements in the Surgical Placement and Loading of Endosseous Implants. J Oral Maxillofac Surg. 2016;74(6):1145-52. doi:10.1016/j.joms.2016.01.048 (opens in a new tab). PMID: 26917203 (opens in a new tab).
  35. Jung GY, Hong I, Müller NPA, Strauss FJ, Lee JS. Clinical agreement between scale-aligned damping capacity and resonance frequency analyses for implant stability: a retrospective method-comparison study. J Periodontal Implant Sci. 2026;56(4):338-350. doi:10.5051/jpis.2505340267 (opens in a new tab). PMID: 41968717 (opens in a new tab).
  36. Barewal RM, Oates TW, Meredith N, Cochran DL. Resonance frequency measurement of implant stability in vivo on implants with a sandblasted and acid-etched surface. Int J Oral Maxillofac Implants. 2003;18(5):641-51. PMID: 14579951 (opens in a new tab).
  37. Oates TW, Valderrama P, Bischof M, Nedir R, Jones A, Simpson J, et al. Enhanced implant stability with a chemically modified SLA surface: a randomized pilot study. Int J Oral Maxillofac Implants. 2007;22(5):755-60. PMID: 17974109 (opens in a new tab).
  38. Abdelraouf SA, Dahab OA, Mostafa B, Kenawy SM, Tawfik OK. Implant stability in the posterior maxilla: clinical and radiographic comparison of osseodensification and conventional drilling: a randomized clinical trial. Clin Oral Investig. 2025;29(10):480. doi:10.1007/s00784-025-06526-8 (opens in a new tab). PMID: 41016918 (opens in a new tab).
  39. Trisi P, Todisco M, Consolo U, Travaglini D. High versus low implant insertion torque: a histologic, histomorphometric, and biomechanical study in the sheep mandible. Int J Oral Maxillofac Implants. 2011;26(4):837-49. PMID: 21841994 (opens in a new tab).
  40. Murakami K, Yamamoto K, Ishida J, Tsutsumi S, Kirita T. Analysis of implant stability changes in immediate loading using a laser displacement sensor in vivo and comparison of its sensitivity with that of resonance frequency analysis. Clin Oral Implants Res. 2021;32(11):1341-1356. doi:10.1111/clr.13835 (opens in a new tab). PMID: 34403162 (opens in a new tab).
  41. Herrero-Climent M, Santos-García R, Jaramillo-Santos R, Romero-Ruiz MM, Fernández-Palacin A, Lázaro-Calvo P, et al. Assessment of Osstell ISQ’s reliability for implant stability measurement: a cross-sectional clinical study. Med Oral Patol Oral Cir Bucal. 2013;18(6):e877-82. doi:10.4317/medoral.19120 (opens in a new tab). PMID: 24121909 (opens in a new tab).
  42. Buyukguclu G, Ozkurt-Kayahan Z, Kazazoglu E. Reliability of the Osstell Implant Stability Quotient and Penguin Resonance Frequency Analysis to Evaluate Implant Stability. Implant Dent. 2018;27(4):429-433. doi:10.1097/ID.0000000000000766 (opens in a new tab). PMID: 29762187 (opens in a new tab).