Skip to main content

Bleeding on Probing at Implants: Diagnostic Validity and Its Consequences for Case Definitions and Treatment Endpoints

Paper type
Critical Review (with a measurement-validity framework)
Status
Original Work · Working Paper
Keywords:
bleeding on probing · peri-implantitis · diagnostic accuracy · case definition · composite endpoint · predictive value · measurement validity
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 (conditional)
  7. 7Limitations
  8. 8Conclusions

Abstract

Background. Bleeding on probing (BoP) is obligatory in the consensus case definition of peri-implant mucositis and is required, together with bone loss, for peri-implantitis. It is also a component of the composite “success” endpoints used in treatment trials. Published accuracy data indicate that dichotomous BoP has high sensitivity but very low specificity for subsequent bone loss, and bleeding varies with probing force, probing depth, probe type, tissue phenotype and access around the prosthesis.

Objective and hypothesis. To quantify what a high-sensitivity, low-specificity sign does when it is made mandatory. We hypothesize that (i) dichotomous BoP under standardized force cannot reach 50% specificity for progressive bone loss; (ii) as a mandatory conjunct it inflates the prevalence of BoP-defined disease and transmits between-study heterogeneity; and (iii) inside a conjunctive composite endpoint it caps the success rate and adds little prognostic discrimination beyond residual probing depth and bone stability.

Proposed approach. A critical appraisal of abstract-level evidence is combined with closed-form and Monte Carlo calculations under stated assumptions (predictive values across a prevalence and accuracy grid; a correlated k-component composite model; site-level versus implant-level aggregation) and a specified prospective prognostic-accuracy design with force-controlled, graded and longitudinal BoP, a bone-loss reference standard, and a sample-size calculation.

Expected contribution. The low specificity of BoP is already published; the contribution is the quantified downstream consequences for definitions and endpoints, a validity framework, numbered falsifiable predictions, and the results that would vindicate BoP as a mandatory criterion. In the illustrative calculations, a component with a 50% pass probability limits a four-component composite to 31-50% success depending on correlation, and the any-of-six-sites rule lowers implant-level specificity far below site-level specificity.

Conclusions. BoP is better supported as a rule-out sign than as a mandatory rule-in criterion. Graded, multi-site or longitudinal bleeding measures deserve prospective testing, and the BoP component of composite endpoints should be validated against later bone loss and implant loss before it is retained.

1. Introduction

Probing is the most common clinical test applied to a dental implant after the prosthesis is delivered. Its main output, bleeding on probing, has been carried over from periodontology, where the absence of bleeding was shown long ago to be a reasonable indicator of stability and the presence of bleeding a poor indicator of progression [1]. At implants the sign has been given a stronger role. The 2017 World Workshop reports require bleeding and/or suppuration on gentle probing for the diagnosis of peri-implant mucositis, and require bleeding and/or suppuration, increasing probing depth and radiographic bone loss for peri-implantitis [2, 3]. Treatment trials then use bleeding again, as one element of a composite definition of success.

The argument of this paper is narrow. A diagnostic sign has two separable properties: how well it detects the condition when present (sensitivity) and how well it stays negative when the condition is absent (specificity). A sign with sensitivity near 1 and specificity near 0 is useful for exclusion and almost useless for confirmation. Making such a sign mandatory has predictable arithmetic consequences, because “mandatory” turns it into a logical conjunct: the case or the success must contain the sign. The consequences depend on where the conjunct is placed. In a definition of disease it governs how many implants are labeled diseased; in a definition of treatment success it governs how many are labeled failures.

Two limits on the claim are stated at the outset. First, the specificity problem is not new. Meta-analyses of the proportion of BoP-positive implants that have peri-implantitis, and longitudinal accuracy cohorts, have said so [4, 5, 6, 7]. Second, in the 2017 peri-implantitis definition BoP is combined with bone loss, so it cannot by itself create a peri-implantitis case; its inflationary effect is confined to mucositis and to any usage in which BoP stands in for disease. The paper therefore does not restate that BoP is imperfect. It asks what follows quantitatively, which endpoint designs are most exposed, and what evidence would settle the matter.

2. Background and critical appraisal of the evidence

Origin of the concept and the periodontal precedent

In a 2.5-year maintenance study of 41 patients with sites reinstrumented only when they bled, frequent bleeding had a sensitivity of 29% and specificity of 88% for attachment loss of 2 mm or more; the positive predictive value was 6% and the negative predictive value 98%, which the authors read as showing that continuous absence of bleeding predicts stability [1]. Systematic reviews of residual probing depth and bleeding as predictors of further attachment and tooth loss [8], and studies of the diagnostic characteristics of clinical tests during supportive therapy [9], belong to the same tradition of evaluating bleeding as a predictive test (their findings were not appraised beyond their titles here). The conceptual origin is therefore a test whose value lies in its negative result. The implant literature inherited the sign but, in the consensus documents, not the caveat.

Accuracy at implants

The best longitudinal evidence is a cohort of 72 patients with 298 implants followed for a mean of 3.9 years, with radiographic bone loss above 1 mm as the reference standard [6]. Bone loss occurred at 9.4% of implants. Baseline BoP had a sensitivity of 96.4% and specificity of 7.4% for later loss; at follow-up, bone loss was always accompanied by BoP (sensitivity 100%, specificity 14.4%). Redness had similar sensitivity and low specificity. Higher thresholds changed the balance: BoP at six sites had a sensitivity of 25.0% and specificity of 88.1%, and suppuration a sensitivity of 14.3% and specificity of 91.5%. The best combined criterion for recent loss was an increase of more than 1 mm in probing depth or in the distance from the prosthesis margin to the soft tissue (sensitivity 82.1%, specificity 70.0%, area under the curve 0.76) [6]. A 24-month supportive-care cohort of 59 patients reached the same qualitative conclusion: 9 of 59 implants (15.3%) showed progressive loss of at least 0.5 mm; BoP and a modified bleeding index were highly sensitive but poorly specific at a single time point, whereas recurrent low-grade bleeding across visits was more strongly associated with progression than isolated profuse bleeding [7]. That cohort was small, with nine events, and was not registered.

Syntheses point the same way. In a systematic review of 31 studies, 24.1% of BoP-positive implants had peri-implantitis (95% confidence interval 19.3-29.7), with a prediction interval of 10.3-69.3 and inconsistent definitions across studies [4]. A later meta-analysis of 30 studies (5,826 patients, 17,198 implants) found 26.5% at the implant level, with a prediction interval of 5.2-56%, and again warned of the false-positive rate [5]. Such wide prediction intervals are what a fixed sign would produce across populations with different prevalences and probing conditions (Section 3).

Contradictory evidence exists. A matched case-control study of 141 patients (1,572 sites at 262 implants) reported specificity of about 90% and sensitivity of 44-53% for sites with mucositis or peri-implantitis against healthy sites [10]. This is the opposite accuracy profile, but the design differs in two ways that favor specificity: cases and controls were selected at the extremes of the clinical spectrum, and the outcome was current disease status rather than later bone loss. Spectrum selection of this kind is known to inflate accuracy estimates.

Determinants of bleeding that are unrelated to disease

Probing force matters. In 17 periodontally maintained patients with healthy peri-implant tissues, raising force by 0.1 N from 0.15 N raised the BoP percentage by 13.7 points at implants and 6.6 points at contralateral teeth; the authors concluded that 0.15 N may be a threshold to avoid false-positive readings [11]. Probing depth matters: in a retrospective multilevel analysis of 1,725 implant sites in 112 patients, the odds of bleeding rose by a factor of 1.6 per millimeter of depth, bleeding probability at 4 mm was 27%, and, importantly, probability at implants did not differ from that at contralateral teeth once depth was controlled [12]. A cross-sectional analysis of 552 sites reported an odds ratio of 1.81 per millimeter [13]. Together these studies indicate that bleeding is a joint function of force, depth and tissue characteristics, with the implant-specific excess partly a depth effect. Whether deeper implant sulci reflect pathology, emergence profile or prosthesis geometry is not resolved.

Prosthesis access adds a further source of variation. Readings taken with the restoration in place agreed exactly with readings after its removal at only 37% of sites in one study of 89 implants, and only the latter correlated with surgically measured bone loss [14]. A recent study found probing 0.89 mm deeper without the prosthesis and higher accuracy of maximum depth, but no significant difference in the best depth-plus-BoP/suppuration combinations (area under the curve 0.709 versus 0.753) [15]. These findings concern depth, but they imply that the sulcular area sampled, and therefore the bleeding obtained, differs with access. Critics have gone further, arguing that probing itself may cause over-diagnosis and that bleeding is a questionable surrogate for implant failure [16], while a commentary has proposed a graded bleeding index to separate traumatic from pathologic induction [17]. A recent review catalogs the diagnostic parameters [18].

Case definitions and prevalence

Prevalence of peri-implantitis ranges from 1% to 47% across studies, with thresholds explaining part of the range [19], and later meta-analyses remain heterogeneous [20, 21]. Within a single Swedish sample the proportion with any peri-implantitis was 45% but only 14.5% for moderate or severe disease [22]. Re-application of the 2017 criteria to an existing cohort roughly halved peri-implantitis prevalence (8.7% of implants and 15.2% of subjects) while mucositis was found at 35.3% of implants and 52.2% of subjects [23]. In a university sample the patient-level prevalence of peri-implantitis was 56.6% under one set of criteria [24]. The secondary (no-baseline) 2017 definition showed high specificity but low sensitivity against documented bone loss in 427 patients examined 9 years after loading [25], and examiner agreement in assigning the 2017 categories to 25 documented implants was only moderate [26]. These findings show that definitions drive prevalence; they do not show that BoP specifically does so, which is the gap this paper addresses.

Composite endpoints in treatment trials

Success defined as probing depth of 5 mm or less, bleeding at no more than one of six sites, no suppuration and no progressive bone loss has been achieved at a minority of implants even when bone levels were stable. In a multicenter trial of reconstructive surgery the primary composite outcome was reached at 14% of implants at 3 years, while a second composite defining disease resolution was reached at 39%, in the presence of probing depth reductions of 3.2-3.5 mm and marginal bone gain of 1.1-1.3 mm at surviving implants [27, 28]. An erythritol air-polishing trial reported success at 19.2% of implants at 1 year with both arms judged to have low success [29]. A trial of adjunctive antibiotics reported 45% success overall, 79% at non-modified and 34% at modified surfaces [30], and a later trial reported 68%, 66% and 28% in its three arms, with bone-level stability of 97%, 89% and 76% [31]. Five-year patient-level success after surgery was 63% but complete resolution, defined as absence of bleeding at all sites, was reached at 42% of implants [32]. A review of long-term outcomes found that the definitions of success varied markedly [33], and a synthesis of long-term surgical studies reported resolution in about 59% and arrest of progressive bone loss in about 70%, with bleeding resolved in 59.9% [34]. The only factor with replicated prognostic value for recurrence is residual deep pocketing: an odds ratio of 7.4 for pockets of 6 mm or more one year after surgery [35], and an approximately eight-fold odds in the synthesis [34].

We found no study that tested whether the BoP component of the composite adds discrimination for later implant loss or progressive bone loss beyond probing depth and bone stability. That absence, not any positive finding, motivates the proposals below.

Table 1. Pivotal studies, with design, size, principal bias and operational evidence level.
Study (ref)Design, nFinding used hereMain bias or limitLevel
Romandini 2025 [6]Cohort, 72 patients, 298 implants, 3.9 yBoP sensitivity 96.4%, specificity 7.4%; BoP at 6 sites specificity 88.1%Single center; 1 mm threshold near radiographic errorB
de Tapia 2026 [7]Prospective cohort, 59 patients, 24 moHigh sensitivity, low specificity of single-visit BoP; repeated low-grade bleeding informative9 events; unregisteredC
Hashim 2018 [4]Systematic review, 31 studies24.1% of BoP-positive implants had peri-implantitis; prediction interval 10.3-69.3Inconsistent definitionsB
Yu 2024 [5]Meta-analysis, 30 studies26.5% at implant levelHeterogeneity; definitionsB
Gerber 2009 [11]Randomized force allocation, 17 patientsBoP rises with force more at implantsSmall, healthy, maintained patientsC
Farina 2017 [12]Retrospective multilevel, 112 patientsDepth drives BoP; implant equals tooth at equal depthRetrospective, single centerB
Monje 2018 [10]Matched case-control, 141 patientsSpecificity about 90%, sensitivity 44-53%Spectrum selectionC
Alibegovic 2025 [28]Randomized trial, 3-year follow-upComposite success 14%; second composite 39%Composite design itselfB
Carcuac 2020 [35]Cohort within trialsResidual pocket of 6 mm or more, odds ratio 7.4Observational prognostic analysisB
Lang 1990 [1]Cohort, 41 periodontal patientsPositive predictive value 6%, negative 98%Teeth; smallB

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

3. Hypothesis and theoretical framework

Formal statements

Let D denote progressive bone loss of a stated amount over a stated interval at an implant, and let T denote a positive BoP result defined by a stated rule (probe, force, number of sites, grade). Let Se = P(T | D), Sp = P(not T | not D) and π = P(D). We state three hypotheses.

H1 (validity). For dichotomous BoP at one or more sites with force of 0.15-0.25 N, Sp for D is below 0.50 in implants under supportive care, whatever reasonable reference threshold between 0.5 and 2 mm is chosen.

H2 (definition). Because the positive predictive value (PPV) of a sign depends on π, a fixed mandatory sign applied to populations with different π and different probing conditions generates between-study heterogeneity in the proportion of positive implants that are truly diseased, even if observers are perfectly reliable.

H3 (endpoint). In a conjunctive composite endpoint, a component with a low pass probability bounds the success rate from above by that probability, and when the component’s failures are mostly false positives the bound is artifactual. Removing the BoP component from the composite does not reduce the discrimination of the 12-month endpoint for later implant loss or progressive bone loss.

Explicit assumptions: (a) the reference standard, radiographic bone change, is measured with error that is small relative to the threshold, or at least non-differential with respect to BoP; (b) bleeding at an implant is partly a function of disease and partly of force, probe, depth, phenotype and access, so that T has a disease-independent component; (c) the clinical purpose of the composite is to detect treatment outcomes that predict later bone loss or implant loss, rather than inflammation as such. Assumption (c) is contestable and is treated as a competing explanation in Section 5.

Predictive values as a function of prevalence

The predictive values follow from Bayes’ theorem:

PPV=(π Se)/(π Se+(1−π)(1−Sp)),  NPV=((1−π) Sp)/((1−π) Sp+π (1−Se))

The apparent proportion of positive implants is πSe+(1−π)(1−Sp). The calculations below were run in the script calc.py (saved in the paper folder; output in calc_output.txt). They are an illustrative computation under stated assumptions, not empirical data. The prevalence values 0.094 and 0.153 are the proportions of implants with progressive bone loss reported in the two longitudinal cohorts [6, 7]; the others are arbitrary grid points. The sensitivity of 0.964 and specificity of 0.074 are the point estimates of the 2025 cohort [6]; the other values are assumptions.

Table 2. Positive predictive value (PPV) and negative predictive value (NPV) for sensitivity of 0.964 across specificity and prevalence of progressive bone loss (computed values).
SpecificityPrev 0.05 PPV/NPVPrev 0.094 PPV/NPVPrev 0.153 PPV/NPVPrev 0.25 PPV/NPVPrev 0.40 PPV/NPV
0.0740.052/0.9750.097/0.9520.158/0.9190.258/0.8600.410/0.755
0.250.063/0.9920.118/0.9850.188/0.9750.300/0.9540.461/0.912
0.500.092/0.9960.167/0.9930.258/0.9870.391/0.9770.562/0.954
0.700.145/0.9970.250/0.9950.367/0.9910.517/0.9830.682/0.967
0.900.337/0.9980.500/0.9960.635/0.9930.763/0.9870.865/0.974

The same script produced grids for sensitivity of 0.90 and 0.75. Four results matter. First, at a specificity of 0.074 and the prevalence of 0.094 the PPV is 0.097, so about nine false positives accompany each true positive (9.26 per true positive), and 93% of implants are BoP-positive. Second, NPV stays above 0.95 for sensitivity 0.964 across the prevalence range except the highest, which supports the rule-out role. Third, the PPV is dominated by prevalence: with Se = 0.964 and Sp = 0.074, PPV reaches 0.5 only at a prevalence of 0.49; with Sp = 0.50 at 0.34; with Sp = 0.90 at 0.094. Fourth, the range of PPV across prevalence 0.05-0.40 at Sp = 0.074 (0.052-0.410) is of the same kind as the spread seen in the published prediction intervals for the proportion of BoP-positive implants with peri-implantitis, 10.3-69.3% [4] and 5.2-56% [5]. The calculation does not reproduce those intervals, which also reflect definitional differences; it shows that a fixed-accuracy sign applied across populations would produce heterogeneity of this order on its own.

As a consistency check on the periodontal precedent, solving the PPV formula for the prevalence consistent with the reported sensitivity of 0.29, specificity of 0.88 and PPV of 0.06 gives π = 0.026 and an NPV of 0.979, close to the reported 98% [1]. This is a derived check, not a reported value.

Where inflation of prevalence arises

For mucositis, BoP is the defining sign, so the label rate equals the apparent positive proportion above, minus any positives excluded by other criteria. For peri-implantitis under the 2017 primary definition (baseline available), BoP is conjoined with bone loss; the number labeled is then bounded by the bone-loss criterion, and a false-positive BoP cannot create a case without bone loss. The effect of BoP on peri-implantitis prevalence is therefore on the sensitivity side: because Se is high, it excludes few true cases. This is why the sign suffers less as a conjunct in the disease definition than in the success definition (Section 3.4), and why the evidence that the secondary definition has high specificity and low sensitivity [25] is a function of the bone and depth thresholds as much as of bleeding. The inflation argument applies with full force in three settings: mucositis prevalence, studies that use BoP-positivity as a proxy for disease, and clinical decisions triggered by any BoP-positive site. Any attribution of the 2017 halving of peri-implantitis prevalence [23] to BoP is unwarranted without a decomposition by criterion.

The k-component composite

Let a composite require components 1 to k to be satisfied, with marginal pass probabilities p_1,…,p_k. The probability of success is bounded by

max(0,∑_i^​ p_i−(k−1)) ≤ P(all) ≤ minip_i.

With independence P=∏p_i. To represent positive correlation we used a Gaussian copula with equicorrelation ρ of latent normal variables: P=Φ_k(Φ^(−1)(p_1),…,Φ^(−1)(p_k);R), estimated by Monte Carlo (2,000,000 draws, seed 40). Assumed marginals for a four-component composite modeled on the common definition: probing depth of 5 mm or less, 0.80; no suppuration, 0.90; no progressive bone loss, 0.85; BoP at no more than one site, p_2 varied. These marginals are assumptions chosen for illustration, not estimates from any trial.

Table 3. Probability that all four components are met (computed; assumed marginals as stated).
BoP component pass probabilityUpper bound min(p)rho = 0rho = 0.3rho = 0.6rho = 0.9
0.900.800.5510.6140.6820.765
0.700.700.4280.5060.5860.674
0.500.500.3060.3810.4470.497
0.300.300.1840.2410.2820.300
0.150.150.0920.1260.1460.149

The independence value for p_2=0.5 matches the closed form (0.8 x 0.5 x 0.9 x 0.85 = 0.306). Two properties follow. A weakly passing component sets a ceiling that no improvement in the other components can exceed: at p_2=0.30 the success rate cannot exceed 0.30 even if bone stability, depth and suppuration were perfect, and the simulated values approach that ceiling as correlation rises. And added components erode success even when each is easy: with all p_i=0.8 and independence, success falls from 0.80 at k = 1 to 0.64, 0.51, 0.41, 0.33 and 0.26 at k = 2 to 6; with equicorrelation 0.5 it falls to 0.69, 0.61, 0.56, 0.51 and 0.48. The reported trial composites (14-45% success with stable bone in most implants) [27, 28, 30, 29] are in the range in which a binding component of this kind operates, but the data do not identify which component binds; that is a testable question.

The consequence for trial interpretation is twofold. Success rates become hard to compare across studies that use different BoP rules, and a treatment effect on bone stability can be diluted: if a treatment raises the bone-stability component without changing the BoP component, the odds ratio on the composite shrinks toward the ratio on the capped component. Both predictions apply only to a conjunctive composite; a hierarchical or graded outcome would behave differently.

Site-level versus implant-level aggregation

Specificity is estimated at the level at which the rule is applied. If a site-level false-positive probability among implants without progression is q, an implant-level rule of “positive if any of six sites bleeds” has false-positive probability 1−(1−q)^6 with independent sites. Sites within an implant are correlated; we therefore modeled the number of bleeding sites X among 6 as beta-binomial with mean q and intra-implant correlation ρ, with a=q(1−ρ)/ρ, b=(1−q)(1−ρ)/ρ and P(X=k)=(6/k)B(k+a,6−k+b)/B(a,b). The values of q are assumptions; the empirical observation that bleeding probability at a 4 mm implant site was 27% in one retrospective analysis [12] shows that site-level rates of this order are plausible for sites with moderate depth, but those sites are not all non-progressing.

Table 4. Implant-level false-positive rate (1 minus specificity) among implants without progression under four aggregation rules (computed; assumed q and intra-implant correlation).
Site-level qCorrelationAt least 1 siteAt least 2 sitesAt least 3 sitesAll 6 sites
0.050.00.2650.0330.0020.0000
0.050.30.1570.0750.0390.0022
0.100.00.4690.1140.0160.0000
0.100.30.2940.1550.0840.0055
0.200.30.5140.3160.1920.0171
0.300.30.6760.4710.3160.0381

With independent sites, a site-level specificity of 0.95 becomes an implant-level specificity of 0.735, and 0.90 becomes 0.531. Correlation within the implant softens this (at q = 0.10 and correlation 0.3 the implant-level false-positive rate is 0.294), but at q = 0.20 it is still 0.514, so implant-level specificity is near 0.49. The aggregation rule matters as much as the sign: requiring at least three sites brings the false-positive rate down to the range 0.002-0.316 depending on q and correlation (and all six sites to below 0.04), while the empirical sensitivity falls sharply (25.0% for six sites in the 2025 cohort) [6]. Small differences in probing force or in depth distribution move the site-level q and are amplified at the implant level. Holding the correlation at 0.3 and Se at 0.964, with prevalence 0.094, implant-level specificity of 0.935, 0.843, 0.706 and 0.486 at q = 0.02, 0.05, 0.10 and 0.20 gives PPVs of 0.604, 0.389, 0.254 and 0.163. A change in q from 0.05 to 0.20, which is the order of the force effect on the BoP percentage in healthy implants [11] (13.7 percentage points per 0.1 N), would therefore reduce PPV by more than half without any change in disease. This comparison is conditional on the assumed parameters.

Numbered, falsifiable predictions

  • P1. In implants under supportive care, with force controlled at 0.15 N, the specificity of dichotomous BoP (at least one of six sites) for bone loss of more than 1 mm over 36 months is below 0.50. Refuted if the lower 95% confidence limit exceeds 0.50.
  • P2. BoP at three or more sites, or a graded index above a threshold, has specificity above 0.70 with sensitivity above 0.50; the area under the curve of a graded or repeated-visit bleeding score exceeds that of dichotomous BoP by at least 0.05. Refuted if the difference is below 0.02 with a confidence interval excluding 0.05.
  • P3. Specificity at 0.25 N is lower than at 0.15 N by at least 0.10 in the same implants. Refuted if the paired difference is under 0.03.
  • P4. In pooled trial data, the 12-month composite without the BoP component discriminates later bone loss above 1 mm or implant loss as well as the composite with it (change in area under the curve below 0.02). Refuted if adding the BoP component raises the area under the curve by 0.05 or more.
  • P5. Residual probing depth of 6 mm or more at 12 months retains an odds ratio of about 5 or higher for later recurrence after adjusting for the BoP component [35, 34].
  • P6. Across trials the heterogeneity in composite success is explained in part by the BoP aggregation rule and probing force; a meta-regression of success on rule explains more heterogeneity than the intervention.
  • P7. Negative predictive value for sustained absence of bleeding exceeds 0.95 at the prevalence observed, supporting a rule-out role [1, 7].

What would vindicate BoP as a mandatory criterion. Specificity with the lower confidence limit above 0.50 under standardized force (P1 refuted); a demonstration that adding the BoP component to depth and bone stability improves discrimination of later implant loss or bone loss by a clinically meaningful margin (P4 refuted) with good calibration; and low between-observer and between-force variability. A result in which BoP contributes only as a rule-out sign would vindicate its use as a screening negative, not as an obligatory criterion.

4. Proposed methodology

Review methods used for this manuscript

The literature was searched on PubMed and Europe PMC on 17 November 2026, and ClinicalTrials.gov was queried for registered studies. Search concepts were diagnostic or predictive accuracy of BoP at implants, probing force, case-definition critiques, composite treatment success, graded bleeding indices, and the periodontal precedent. Titles and abstracts were read; full texts were not available and no risk-of-bias tool was applied beyond the design notes in Table 1. Embase, Scopus and Web of Science were not accessible, so the review is not systematic and may have missed studies. Two registered observational studies were identified: NCT07004517 (peri-implant probing accuracy, linked to the study in [15]) and NCT03031392 (sensitivity and specificity of peri-implantitis parameters; no results located).

Study A: prospective prognostic-accuracy study

Design and population. A multicenter prospective cohort of adults with bone-level or tissue-level implants in function for at least 12 months, enrolled at a supportive-care visit with a standardized baseline radiograph, excluding implants with already established peri-implantitis requiring treatment. The unit of analysis is the implant; patient is the cluster. Sites are analyzed secondarily.

Index tests (all recorded at each visit, at 0, 12, 24 and 36 months). (i) Dichotomous BoP at six sites with a pressure-limited probe calibrated to 0.15 N and recorded 30 seconds after probing. (ii) The same sites with a 0.25 N probe in a randomized split-sequence, so that the force effect is estimated within implants (paired, randomized order, with an interval long enough for the tissue to settle). (iii) A graded bleeding score (none, pinpoint, line, profuse) at each site. (iv) Summary rules: at least 1, at least 2, at least 3 and all 6 sites; count of bleeding sites; and the pattern across visits (never, once, repeated). (v) Visual redness and swelling, suppuration. (vi) Covariates: probing depth, probe geometry and tip diameter, keratinized tissue width, mucosal phenotype, prosthesis access (probing with restoration in place, and after removal where feasible and clinically indicated), emergence profile, plaque, smoking, and prior periodontitis.

Reference standard. Progressive bone loss, defined as radiographic marginal bone-level change from baseline to month 36 (primary) of more than 1 mm from standardized, calibrated radiographs read by two blinded readers with adjudication; thresholds of 0.5 mm and 2 mm are secondary. Because measurement error near 1 mm is not negligible, the analysis includes a repeat-measurement subsample to estimate the reproducibility of bone change and propagate it as non-differential misclassification of the reference standard. Implant loss and any treatment prompted by the clinician are recorded as secondary outcomes, and a clinical-use decision such as retreatment is recorded to detect incorporation bias, since clinicians who see bleeding may intervene.

Blinding and bias control. Radiographic readers are blinded to clinical data; clinicians are blinded to the radiographic reading of bone change at each visit (radiographs taken but not evaluated until the end) when ethically feasible, with a safety rule that any visible severe loss triggers review. Examiners are calibrated, and probing-force devices are verified.

Ethics and feasibility. The force-comparison arm adds a second probing pass to routine care. Ethics approval and prospective registration would be required and are not claimed.

Analysis plan

For each rule, sensitivity, specificity, PPV and NPV and likelihood ratios are estimated with cluster-robust (patient-level) confidence intervals; the primary contrast is specificity of at least 1-of-6 dichotomous BoP at 0.15 N against the threshold 0.50. Graded and repeated-visit measures are compared with dichotomous BoP by the paired difference in area under the receiver-operating curve, with bootstrap intervals resampling patients. Calibration is assessed for a multivariable model of depth, bone level and bleeding variables. Decision-curve analysis compares strategies (treat if any BoP; treat if multi-site BoP; follow-up only) across plausible thresholds, because the value of a false positive depends on the intervention it triggers. The effect of force, probe, phenotype and prosthesis access on false-positive rates is estimated with a mixed-effects logistic model with random effects for implant and patient. Prevalence-dependent quantities are reported across a prevalence grid.

Sample size. For comparing a proportion with a reference value, with p_0 the null and p_1 the alternative:

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

multiplied by the design effect 1+(m−1)ρ, with m implants per patient and ρ the intra-patient correlation. The assumptions are m = 1.8, ρ = 0.3 (design effect 1.24), one-sided α = 0.05 and power 0.80. For specificity, to detect Sp=0.65 against the null 0.50 requires 67 implants without progression, or 83 after the design effect. For sensitivity, to detect 0.92 against 0.80 requires 55 progressing implants (68 after design effect). At a progression prevalence of 0.12 (between the two cohorts’ 0.094 and 0.153 [6, 7]) these requirements imply 567 implants in about 315 patients, or 668 implants in about 371 patients allowing for 15% attrition (about 80 progressing implants). If instead precision is the objective, estimating Sp = 0.70 with a half-width of 0.06 requires 225 implants without progression (278 after the design effect), and estimating Se = 0.90 with a half-width of 0.07 requires 71 progressing implants (88 after the design effect); at the same prevalence this implies 734 implants in about 408 patients, or 864 implants and 480 patients with 15% attrition. The precision target is the more conservative plan and is the recommended one. All arithmetic was computed in Python (calc.py, calc2.py).

Study B: individual-patient re-analysis of trial composites

Pooled individual-patient data from trials of peri-implantitis treatment with at least 3 years of follow-up would be used to compare the discrimination of 12-month endpoints for later bone loss above 1 mm or implant loss: (a) probing depth of 5 mm or less plus bone stability; (b) the same plus no suppuration; (c) the full composite including BoP at no more than one site; (d) a graded alternative. Metrics are area under the curve with paired comparison, net reclassification, and calibration. A decomposition of each trial’s failure to meet the composite by component, together with the correlation among component failures, would show which component binds and would test the capping model in Section 3.4 directly. If individual-patient data cannot be obtained, aggregate component-level data can still test P6 by meta-regression.

Study C: re-application of definitions

Where baseline radiographs exist, the primary and secondary 2017 case definitions would be applied to several cohorts, then re-applied with BoP replaced by multi-site, graded or longitudinal bleeding. The quantities of interest are prevalence ratios, and sensitivity and specificity for documented bone loss, with a decomposition by criterion so that the contribution of BoP to the change in prevalence can be isolated. This addresses the gap noted in Section 3.3.

5. Competing explanations and how the design discriminates among them

Low specificity reflects reference-standard failure rather than a poor sign. Bone loss accrues slowly and is measured with error; implants with persistent inflammation but no loss within 36 months would be counted as false positives. The design addresses this by using several thresholds, a longer window in a subset, reproducibility estimation, and by reporting implant loss and treatment received. If specificity remains low across all thresholds and the extended window, the reference-standard explanation is weakened; if it rises with the window, the sign has prognostic value on a longer horizon than 36 months.

Low specificity reflects disease-independent bleeding (force, depth, phenotype, access). The paired force arm, covariates and mixed model estimate this component directly. If specificity at 0.15 N is similar to that at 0.25 N and to unadjusted estimates, the force explanation is not supported; the published difference between 0.15 and 0.25 N in healthy implants [11] would then be of limited importance for the diagnostic problem.

Bleeding is valid as a marker of inflammation, and the target (bone loss) is the wrong comparator. This is the strongest alternative. Mucositis is defined by inflammation, and bleeding can be an appropriate measure of it without predicting bone loss; the argument against mandatory use does not apply to that purpose. The design cannot settle a definitional question, but it separates the two purposes: accuracy for inflammation would need a histologic or biomarker reference, which is outside Study A, while Study A tests prognosis. The conclusion is therefore restricted: BoP may be valid as a descriptor of inflammation yet invalid as a prognostic rule-in criterion.

Implants are no different from teeth. Bleeding probability at implants was similar to that at contralateral teeth when depth was controlled [12], so an implant-specific false-positive problem may be a depth problem. Study A includes contralateral tooth sites for a subset to test whether the false-positive structure is shared. If the same low specificity applies to teeth, the problem is the sign in general (consistent with the periodontal precedent [1]) rather than an implant artifact.

Multi-site and longitudinal patterns may mitigate the problem. The 2025 cohort and the 2026 cohort both point in this direction, although in the 2025 cohort bleeding at six sites had low sensitivity [6, 7].

6. Clinical and engineering implications (conditional)

If P1 and P4 are supported, several changes would be reasonable. Case definitions could retain BoP as a descriptor of mucositis while labeling it as a non-prognostic sign, and could require a graded or multi-site threshold when the purpose is to trigger intervention. Trial protocols could report the composite by component and report bone-level stability and implant survival as separate primary outcomes, with the BoP component either excluded from the primary composite or analyzed as a graded secondary outcome. Pre-specified sample sizes for trials that use the full composite should account for the ceiling set by the lowest component pass probability (Section 3.4). Clinically, a negative BoP result could be used as a reason to defer intervention, consistent with the periodontal precedent [1], while a positive single-site result would be considered insufficient for treatment decisions without bone, depth and trend information.

If P2 is supported, a graded index or longitudinal pattern could become the standard measurement, with force-controlled probing. Probe design and force calibration are established technologies, and no new device is implied; the engineering task is standardizing force, tip geometry and timing across studies [11, 17].

7. Limitations

The calculations rest on assumed parameters. The marginal pass probabilities in Table 3 were not estimated from any trial; the site-level false-positive probabilities and the correlation in Table 4 were not measured; the Gaussian-copula and beta-binomial forms are modeling conveniences. The sensitivity and specificity values come from a single cohort of 72 patients [6], with modest event numbers, and the 2026 cohort had nine events [7]. The low-specificity conclusion therefore rests on few primary datasets and on radiographic loss over a few years as the reference.

The appraisal was at abstract level; reported figures were not checked against full texts, and subgroup or threshold analyses in the full texts may qualify them. Searches were limited to PubMed, Europe PMC and ClinicalTrials.gov, with several databases unavailable. We did not find a study testing the prognostic contribution of the BoP component of trial composites; a study may exist outside the searched sources.

Radiographic bone change has limited resolution and does not detect buccal or lingual loss; bone loss in the first year after loading is partly remodeling. The prevalence of progression in supportive-care cohorts may not apply to implants with established disease, for which the PPV of BoP would be higher (Table 2). Finally, bleeding resolution is a legitimate marker of inflammatory control; the argument concerns its role as a mandatory conjunct, not its use as a measured outcome.

8. Conclusions

The data reviewed support a limited and specific conclusion. Bleeding on probing at implants is sensitive but, as a dichotomous single-time-point sign, appears poorly specific for subsequent bone loss, and it varies with factors unrelated to disease. The quantitative consequences, under stated assumptions, are that predictive values are dominated by prevalence, that a conjunctive composite cannot exceed the pass probability of its weakest component, and that an any-site aggregation rule can reduce implant-level specificity far below site-level specificity. These calculations are illustrative, but they show why a mandatory BoP rule would produce heterogeneity between studies and low apparent success even when bone is stable. The argument is weaker for peri-implantitis definitions, in which BoP is combined with bone loss, than for mucositis and composite endpoints. The negative predictive value of sustained absence of bleeding supports its use as a rule-out sign. Whether any BoP-based criterion deserves mandatory status can be settled by the prospective design and the trial re-analysis specified here, with predictions that can fail.

Cite this paper

Jain S, Gupta A. Bleeding on Probing at Implants: Diagnostic Validity and Its Consequences for Case Definitions and Treatment Endpoints. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/bleeding-on-probing-validity-endpoints

References

  1. Lang NP, Adler R, Joss A, Nyman S. Absence of bleeding on probing. An indicator of periodontal stability. J Clin Periodontol. 1990;17(10):714-21. doi:10.1111/j.1600-051x.1990.tb01059.x (opens in a new tab). PMID: 2262585 (opens in a new tab).
  2. Berglundh T, Armitage G, Araujo MG, Avila-Ortiz G, Blanco J, Camargo PM, et al. Peri-implant diseases and conditions: Consensus report of workgroup 4 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Clin Periodontol. 2018;45 Suppl 20:S286-S291. doi:10.1111/jcpe.12957 (opens in a new tab). PMID: 29926491 (opens in a new tab).
  3. Renvert S, Persson GR, Pirih FQ, Camargo PM. Peri-implant health, peri-implant mucositis, and peri-implantitis: Case definitions and diagnostic considerations. J Clin Periodontol. 2018;45 Suppl 20:S278-S285. doi:10.1111/jcpe.12956 (opens in a new tab). PMID: 29926496 (opens in a new tab).
  4. Hashim D, Cionca N, Combescure C, Mombelli A. The diagnosis of peri-implantitis: A systematic review on the predictive value of bleeding on probing. Clin Oral Implants Res. 2018;29 Suppl 16:276-293. doi:10.1111/clr.13127 (opens in a new tab). PMID: 30328188 (opens in a new tab).
  5. Yu X, Lin X, Wang F, Wu Y. LONG-TERM PREDICTIVE VALUE OF BLEEDING ON PROBING IN PERI‑IMPLANTITIS DIAGNOSIS: A SYSTEMATIC REVIEW AND META-ANALYSIS. J Evid Based Dent Pract. 2024;24(4):102034. doi:10.1016/j.jebdp.2024.102034 (opens in a new tab). PMID: 39631968 (opens in a new tab).
  6. Romandini M, Lima C, Moreno M, Sanz M. Accuracy of Clinical Parameters in Predicting/Diagnosing Peri-Implant Bone Loss. J Clin Periodontol. 2025;52(8):1070-1081. doi:10.1111/jcpe.14095 (opens in a new tab). PMID: 40320760 (opens in a new tab).
  7. de Tapia B, Torrens G, di Leone F, Amerio E, Valles C, Nart J, et al. Bleeding on Probing as a Predictor of Peri-Implant Bone Loss During Supportive Care: A Prospective Cohort Study. Clin Implant Dent Relat Res. 2026;28(2):e70138. doi:10.1111/cid.70138 (opens in a new tab). PMID: 41873754 (opens in a new tab).
  8. Renvert S, Persson GR. A systematic review on the use of residual probing depth, bleeding on probing and furcation status following initial periodontal therapy to predict further attachment and tooth loss. J Clin Periodontol. 2002;29 Suppl 3:82-9; discussion 90-1. doi:10.1034/j.1600-051x.29.s-3.2.x (opens in a new tab). PMID: 12787209 (opens in a new tab).
  9. Luterbacher S, Mayfield L, Brägger U, Lang NP. Diagnostic characteristics of clinical and microbiological tests for monitoring periodontal and peri-implant mucosal tissue conditions during supportive periodontal therapy (SPT). Clin Oral Implants Res. 2000;11(6):521-9. doi:10.1034/j.1600-0501.2000.011006521.x (opens in a new tab). PMID: 11168245 (opens in a new tab).
  10. Monje A, Caballé-Serrano J, Nart J, Peñarrocha D, Wang HL, Rakic M. Diagnostic accuracy of clinical parameters to monitor peri-implant conditions: A matched case-control study. J Periodontol. 2018;89(4):407-417. doi:10.1002/JPER.17-0454 (opens in a new tab). PMID: 29683503 (opens in a new tab).
  11. Gerber JA, Tan WC, Balmer TE, Salvi GE, Lang NP. Bleeding on probing and pocket probing depth in relation to probing pressure and mucosal health around oral implants. Clin Oral Implants Res. 2009;20(1):75-8. doi:10.1111/j.1600-0501.2008.01601.x (opens in a new tab). PMID: 19126110 (opens in a new tab).
  12. Farina R, Filippi M, Brazzioli J, Tomasi C, Trombelli L. Bleeding on probing around dental implants: a retrospective study of associated factors. J Clin Periodontol. 2017;44(1):115-122. doi:10.1111/jcpe.12647 (opens in a new tab). PMID: 27943375 (opens in a new tab).
  13. Merli M, Bernardelli F, Giulianelli E, Toselli I, Mariotti G, Nieri M. Peri-implant bleeding on probing: a cross-sectional multilevel analysis of associated factors. Clin Oral Implants Res. 2017;28(11):1401-1405. doi:10.1111/clr.13001 (opens in a new tab). PMID: 28165160 (opens in a new tab).
  14. Serino G, Turri A, Lang NP. Probing at implants with peri-implantitis and its relation to clinical peri-implant bone loss. Clin Oral Implants Res. 2013;24(1):91-5. doi:10.1111/j.1600-0501.2012.02470.x (opens in a new tab). PMID: 22462625 (opens in a new tab).
  15. Xu M, Bao S, Qian Y, Qiu C, Zhang J, Si M. Diagnostic Accuracy of Peri-Implant Probing With or Without Prostheses in Detecting Peri-Implantitis. Clin Oral Implants Res. 2026;37(5):612-622. doi:10.1111/clr.70104 (opens in a new tab). PMID: 41708887 (opens in a new tab).
  16. Coli P, Sennerby L. Is Peri-Implant Probing Causing Over-Diagnosis and Over-Treatment of Dental Implants?. J Clin Med. 2019;8(8). doi:10.3390/jcm8081123 (opens in a new tab). PMID: 31362381 (opens in a new tab).
  17. Dukka H, Saleh MHA, Ravidà A, Greenwell H, Wang HL. Is bleeding on probing a reliable clinical indicator of peri-implant diseases?. J Periodontol. 2021;92(12):1669-1674. doi:10.1002/JPER.20-0890 (opens in a new tab). PMID: 33829501 (opens in a new tab).
  18. Monje A, Salvi GE. Diagnostic methods/parameters to monitor peri-implant conditions. Periodontol 2000. 2024;95(1):20-39. doi:10.1111/prd.12584 (opens in a new tab). PMID: 38923148 (opens in a new tab).
  19. Derks J, Tomasi C. Peri-implant health and disease. A systematic review of current epidemiology. J Clin Periodontol. 2015;42 Suppl 16:S158-71. doi:10.1111/jcpe.12334 (opens in a new tab). PMID: 25495683 (opens in a new tab).
  20. Rakic M, Galindo-Moreno P, Monje A, Radovanovic S, Wang HL, Cochran D, et al. How frequent does peri-implantitis occur? A systematic review and meta-analysis. Clin Oral Investig. 2018;22(4):1805-1816. doi:10.1007/s00784-017-2276-y (opens in a new tab). PMID: 29218422 (opens in a new tab).
  21. Diaz P, Gonzalo E, Villagra LJG, Miegimolle B, Suarez MJ. What is the prevalence of peri-implantitis? A systematic review and meta-analysis. BMC Oral Health. 2022;22(1):449. doi:10.1186/s12903-022-02493-8 (opens in a new tab). PMID: 36261829 (opens in a new tab).
  22. Derks J, Schaller D, Håkansson J, Wennström JL, Tomasi C, Berglundh T. Effectiveness of Implant Therapy Analyzed in a Swedish Population: Prevalence of Peri-implantitis. J Dent Res. 2016;95(1):43-9. doi:10.1177/0022034515608832 (opens in a new tab). PMID: 26701919 (opens in a new tab).
  23. Shimchuk AA, Weinstein BF, Daubert DM. The impact of a change in classification criteria on the prevalence of peri-implantitis: A cross-sectional analysis. J Periodontol. 2021;92(9):1339-1346. doi:10.1002/JPER.20-0566 (opens in a new tab). PMID: 33277941 (opens in a new tab).
  24. Romandini M, Lima C, Pedrinaci I, Araoz A, Soldini MC, Sanz M. Prevalence and risk/protective indicators of peri-implant diseases: A university-representative cross-sectional study. Clin Oral Implants Res. 2021;32(1):112-122. doi:10.1111/clr.13684 (opens in a new tab). PMID: 33210772 (opens in a new tab).
  25. Romandini M, Berglundh J, Derks J, Sanz M, Berglundh T. Diagnosis of peri-implantitis in the absence of baseline data: A diagnostic accuracy study. Clin Oral Implants Res. 2021;32(3):297-313. doi:10.1111/clr.13700 (opens in a new tab). PMID: 33340418 (opens in a new tab).
  26. Marini L, Tomasi C, Gianserra R, Graziani F, Landi L, Merli M, et al. Reliability assessment of the 2018 classification case definitions of peri-implant health, peri-implant mucositis, and peri-implantitis. J Periodontol. 2023;94(12):1461-1474. doi:10.1002/JPER.23-0129 (opens in a new tab). PMID: 37322858 (opens in a new tab).
  27. Derks J, Ortiz-Vigón A, Guerrero A, Donati M, Bressan E, Ghensi P, et al. Reconstructive surgical therapy of peri-implantitis: A multicenter randomized controlled clinical trial. Clin Oral Implants Res. 2022;33(9):921-944. doi:10.1111/clr.13972 (opens in a new tab). PMID: 35804491 (opens in a new tab).
  28. Alibegovic L, Trullenque-Eriksson A, Ortiz-Vigón A, Guerrero A, Donati M, Bressan E, et al. Bone Substitute Material in the Surgical Therapy of Peri-Implantitis-3-Year Outcomes of a Randomized Controlled Trial. Clin Oral Implants Res. 2025;36(4):436-448. doi:10.1111/clr.14393 (opens in a new tab). PMID: 39704200 (opens in a new tab).
  29. Hentenaar DFM, De Waal YCM, Stewart RE, Van Winkelhoff AJ, Meijer HJA, Raghoebar GM. Erythritol air polishing in the surgical treatment of peri-implantitis: A randomized controlled trial. Clin Oral Implants Res. 2022;33(2):184-196. doi:10.1111/clr.13881 (opens in a new tab). PMID: 34808006 (opens in a new tab).
  30. Carcuac O, Derks J, Charalampakis G, Abrahamsson I, Wennström J, Berglundh T. Adjunctive Systemic and Local Antimicrobial Therapy in the Surgical Treatment of Peri-implantitis: A Randomized Controlled Clinical Trial. J Dent Res. 2016;95(1):50-7. doi:10.1177/0022034515601961 (opens in a new tab). PMID: 26285807 (opens in a new tab).
  31. Riben Grundström C, Lund B, Kämpe J, Belibasakis GN, Hultin M. Systemic antibiotics in the surgical treatment of peri-implantitis: A randomized placebo-controlled trial. J Clin Periodontol. 2024;51(8):981-996. doi:10.1111/jcpe.13994 (opens in a new tab). PMID: 38699828 (opens in a new tab).
  32. Heitz-Mayfield LJA, Salvi GE, Mombelli A, Loup PJ, Heitz F, Kruger E, et al. Supportive peri-implant therapy following anti-infective surgical peri-implantitis treatment: 5-year survival and success. Clin Oral Implants Res. 2018;29(1):1-6. doi:10.1111/clr.12910 (opens in a new tab). PMID: 27335316 (opens in a new tab).
  33. Roccuzzo M, Layton DM, Roccuzzo A, Heitz-Mayfield LJ. Clinical outcomes of peri-implantitis treatment and supportive care: A systematic review. Clin Oral Implants Res. 2018;29 Suppl 16:331-350. doi:10.1111/clr.13287 (opens in a new tab). PMID: 30328195 (opens in a new tab).
  34. Monje A, Pons R, Ramanauskaite A, Castro A, Schwarz F, Chambrone L. Long-term surgical treatment outcomes of peri-implantitis. Periodontol 2000. 2025. doi:10.1111/prd.12643 (opens in a new tab). PMID: 40693702 (opens in a new tab).
  35. Carcuac O, Derks J, Abrahamsson I, Wennström JL, Berglundh T. Risk for recurrence of disease following surgical therapy of peri-implantitis-A prospective longitudinal study. Clin Oral Implants Res. 2020;31(11):1072-1077. doi:10.1111/clr.13653 (opens in a new tab). PMID: 32870513 (opens in a new tab).