Continuous local antibiotic perfusion for orthopaedic infections: A systematic review and pooled individual participant data analysis of observational reports

Abstract

Background

Orthopaedic infections are difficult to eradicate because biofilm and poor local vascularity limit antibiotic exposure. Continuous local antibiotic perfusion (CLAP) delivers sustained, titratable antibiotics directly into infected compartments. We used harmonised individual participant data (IPD) to quantify early effectiveness, longer-term control, safety, and patient-level modifiers.

Methods

We performed an IPD review of observational reports using CLAP as primary or adjunctive therapy (January–May 2025). The primary outcome was 30-days early response (C-reactive protein ≤3 mg/L or earliest sustained clinical/wound improvement). Secondary outcomes were durable infection control at ≥6 and ≥12 months using evaluable denominators with best–worst bounds, infection-free days) and safety. One-stage analyses used mixed-effects logistic regression; Restricted Mean Survival Time (RMST) was preferred when proportional hazards were violated. Multiple imputation supported inferences.

Results

Eighty-one studies (n = 256) were included; 164 patients had observed time-to-response. Fifty-nine percent achieved a 30-days response; median time-to-response was 26 days. Implant involvement was associated with lower odds of 30-days response; trajectories were slower with implants and higher organism burden (polymicrobial ≥3), while osteomyelitis responded faster than fracture-related infection. RMST (30) showed delays with implants (+4.43 days) and polymicrobial infection (+6.74 days), and faster response for osteomyelitis versus fracture-related infection (−9.06 days). Durable control among evaluable patients was 88.4% at ≥6 months and 90.2% at ≥12 months, with best–worst bounds of 89.2–82.2% and 90.9–83.5%, respectively. Infection-free-day RMST supported substantial time free of recurrent infection within the first year. Adverse events were uncommon; renal events were generally reversible.

Conclusions

CLAP achieved encouraging early response and high durability among evaluable patients, with slower trajectories when implants were retained or pathogen burden was high and faster responses in osteomyelitis. Safety appeared acceptable with monitoring. Prospective comparative studies using standardised endpoints, with RMST for non-proportional hazards, are warranted.

Keywords

continuous local antibiotic perfusion osteomyelitis arthritis infectious prosthesis-related infections surgical wound infection biofilms

Introduction

Orthopaedic infections—including osteomyelitis, septic arthritis, fracture-related infection (FRI), and periprosthetic joint infection (PJI)—remain difficult to eradicate and often require repeated debridement, prolonged systemic antibiotics, and complex implant decisions. The burden is substantial and rising with population ageing, higher arthroplasty volumes, complex trauma, and multimorbidity. Population-level data show a sustained and projected increase in musculoskeletal infectious morbidity and related service demand.^1,2 Even with contemporary strategies such as debridement, antibiotics and implant retention (DAIR), outcomes are limited by biofilm biology and compromised local vascularity, which reduce effective concentration–time exposure at the nidus and permit persistent or recrudescent infection.^3–6

These constraints motivate local strategies that deliver high antimicrobial exposure in situ while minimising systemic toxicity.⁷ Continuous local antibiotic perfusion (CLAP) aims to meet that need: catheters are placed within the infected bone, joint, or soft tissues to allow continuous, titratable antibiotic ingress to the target space.^7,8 Pharmacologically, prolonged high local exposure can increase biofilm susceptibility and overcome time-dependent tolerance.^9,10 Critically, CLAP is typically designed with controlled egress by pairing perfusion with negative-pressure wound therapy (NPWT) or dedicated drainage, which helps avoid pooling, promotes clearance of exudate, and enables monitoring of effluent; this ingress–egress circuit distinguishes CLAP from instillation-only approaches.^11–13

Despite these practical advantages, the evidence base is largely case reports and small series with heterogeneity in techniques, denominators, and outcome definitions. An individual-participant-data (IPD) approach allows harmonised diagnostic frameworks, uniform outcome timing/definitions, and prespecified exploration of patient-level modifiers while accounting for study-level heterogeneity.¹⁴ Because early decision points matter clinically, we prioritised a 30-days early-response endpoint (C-Reactive Protein (CRP) ≤3 mg/L or earliest sustained clinical/wound improvement when CRP is unavailable), supported by laboratory reference ranges and prior CLAP series using similar thresholds.^15–18 A 30-days horizon is also aligned with surgical site infection (SSI) surveillance windows, trial feasibility, and health-system quality metrics (e.g., 30-days readmission).^19–23 Longer-term control (“durability”) was evaluated at prespecified windows with transparent denominators and bounds.

We sought to (i) quantify the proportion achieving early response by 30 days and characterise the distribution of time-to-early-response; (ii) estimate longer-term control at prespecified windows using clear evaluable denominators and best–worst bounds; (iii) explore prespecified patient-level modifiers (implant involvement, organism groups, organism counts, infection type/site/phase, selected comorbidities) for the primary endpoint within a one-stage IPD framework; and (iv) summarise recurrence/relapse and complications (major/minor, including antibiotic-related adverse events). Protocol features (site of antibiotic perfusion, infusion duration/dose/concentration) were not primary analytic targets.

Methods

Protocol and reporting standards

We followed the Preferred Reporting Items for Systematic Review–Individual Participant Data (PRISMA-IPD) guidelines¹⁴ and PRISMA 2020 (Table S1).²⁴ The protocol was registered with PROSPERO (CRD42025635194; 21 Jan 2025). Two prospectively recorded amendments (25 Mar 2025; pre-extraction, and 20 September 2025) formalized IPD methods and clarified the outcome hierarchy (primary: 30-days early response; secondary: durable infection control and infection-free days).

Patient cohort and eligibility criteria

We included case reports, case series, and cohort studies reporting continuous local antibiotic perfusion (CLAP) as part of an infection management pathway in orthopaedic settings. CLAP was defined as continuous intralesional antibiotic ingress via catheters into bone, joint, or soft tissues with controlled egress through negative-pressure wound therapy (NPWT) or dedicated drainage/aspiration. Variants included intramedullary (iMAP), intra–soft-tissue (iSAP), and intra-articular (iJAP) approaches. We excluded studies using intermittent instillation without controlled egress and studies focused only on other local carriers (e.g., beads/spacers) without continuous perfusion. Populations included adults and children with osteomyelitis, periprosthetic joint infection (PJI), fracture-related infection (FRI), surgical-site infection involving musculoskeletal compartments, or septic arthritis. Non-orthopaedic infections, animal models, and in-vitro reports were excluded. Screening occurred at the study level; during IPD extraction, any individual patients who did not actually receive CLAP were removed while retaining eligible cases from the same report.

Outcomes and definitions

Primary outcome

Early response within 30 days of CLAP initiation, defined as C-reactive protein (CRP) ≤3 mg/L or, where CRP was unavailable, the earliest sustained clinical or wound improvement within 30 days^15–18 The binary indicator was derived from observed response times when available; cases described only as ‘response by 30 days’ without an exact response time contributed to the 30-days endpoint but were treated as censored in time-to-event analyses.

Secondary outcomes

(i) Durability of infection control at prespecified windows (≥6 and ≥12 months) using evaluable denominators (patients with follow-up at/after the window and determinate status), with best–worst bounds for unknowns. By “best–worst bounds” we mean the best case assigns all unknown-status patients to success and the worst-case assigns all to failure, yielding a transparent range when follow-up is incomplete. (ii) Infection-free days (IFD) up to fixed horizons (τ = 180 and 365 days) via restricted mean survival time (RMST) using interval-censored methods. (iii) Safety, including major/minor complications and antibiotic-related adverse events (renal, otologic). (iv) Recurrence/relapse and infection-related mortality. (v) Functional recovery when available (descriptive).

A patient was counted as having their infection durably controlled if, by last observation at/after the window, there was no documented recurrence/relapse and no infection-related death. Patients with follow-up at/after the window but indeterminate status (e.g., missing recurrence documentation) were treated as unknown and excluded from the evaluable denominator; we report best–worst bounds by assigning all unknowns to success or failure. In a strict sensitivity definition, durable control additionally required no re-operation for infection and no chronic suppressive antibiotics (CSA) when explicitly reported; lack of reporting was not penalized.

Rationale for the 30-days endpoint

The first month captures the decisive postoperative window when inflammatory markers and wound status typically declare trajectory and when escalation or de-escalation decisions are made. Within our IPD, the 30-days mark lay between the cohort’s median and upper-quartile treatment response times, supporting a threshold that reflects meaningful early treatment response without being overly permissive. Because any single threshold can obscure distributional detail, we paired the binary 30-days analysis with time-to-event and restricted mean survival time (RMST) approaches, preserving information about the full clearance-time distribution and enabling interpretation in absolute days gained or lost over a fixed horizon.

Information sources and search strategy

We searched MEDLINE (PubMed and Ovid), EMBASE (ProQuest), Scopus, J-STAGE, Isho. jp Web, CiNii Research, and the Cochrane Library from inception to 1 May 2025, without language or publication-status restrictions. Trial registries (ClinicalTrials.gov, WHO ICTRP) were searched but yielded no eligible records. We screened reference lists of included studies and prior reviews, scanned citation indices, and searched grey literature (Google Scholar, ResearchGate). Authors were contacted for restricted-access publications and to clarify eligibility or data elements. Database search start dates were: PubMed 10 Jan 2025; Scopus 10 Jan 2025; Ovid MEDLINE 11 Jan 2025; J-STAGE 11 Jan 2025; Cochrane Library 12 Jan 2025; EMBASE (ProQuest) 12 Jan 2025; Isho. jp Web and CiNii Research 13 Jan 2025. Full strategies (one verbatim, others adapted) are in the Supplemental Appendix.

Study selection, data collection, and management

Two reviewers screened titles/abstracts and full texts independently; disagreements were resolved by discussion or a third reviewer. To minimise assessor bias, screening and IPD extraction/harmonisation were performed independently using a prespecified codebook; disagreements were resolved by consensus or third-party adjudication. Where feasible, reconstructed IPD were cross-checked against source tables/figures and verified via author queries. Overlapping series were checked for duplicate patients by centre, time frame, and key characteristics; duplicates were removed. For eligible studies, we requested individual participant data (IPD) covering demographics; infection type, anatomical site, and phase; implant management; microbiology; comorbidities; CLAP modality and antibiotic details (drug, concentration, dose, flow/dwell, duration; NPWT parameters); biomarkers; clinical course; follow-up; and outcomes. Where author-provided IPD was unavailable, we reconstructed patient-level data from published tables/figures/narratives when feasible. Study-level variables (country, centre, design, year, sample size, follow-up) were recorded. Data were stored in a secure repository (Google Drive) with restricted access. Integrity checks included range and plausibility limits for continuous variables; temporal logic checks (procedure → response/failure → last follow-up); and cross-field consistency (e.g., infection type vs anatomical site). Values were reconciled against published tables/figures/narratives; queries to authors resolved discrepancies where possible.

Study-versus patient-level eligibility

Screening at the study level maximized capture of CLAP cohorts across designs and languages. During IPD harmonization, we then applied patient-level verification: individuals misclassified as receiving CLAP or treated exclusively with other local strategies were excluded while eligible co-patients from the same study were retained. This two-step approach reduces selection bias from excluding otherwise informative reports, preserves per-study clustering for one-stage modelling, and ensures that effectiveness estimates are derived strictly from participants who actually received CLAP. It also maintains fidelity to the real-world heterogeneity of CLAP deployment while allowing standardized outcome definitions and covariate structures at the individual level.

Risk of bias assessments

Within studies, we used the CARE Case Report Checklist and JBI Case Report/Case Series Critical Appraisal Tool, which are presented as study-level summaries and traffic-light plots (Figure 1).^25,26 IPD integrity checks (sequence plausibility, completeness, cross-field consistency) corroborated these ratings. Across studies, small-study effects/publication bias were evaluated visually (funnel plots) and statistically via Egger’s regression (p < 0.05 indicating asymmetry), which results (interpret cautiously) are provided in the Supplementary Appendix (Figures S1 and S2).²⁷

Figure 1.

Risk-of-bias assessments utilizing CARE and JBI guidelines: (A) case reports and (B) case series, displayed as traffic-light plots. Abbreviations: CARE: CAse REport guidelines; JBI: Joanna Briggs institute.

Clinical and microbiological classifications

Infection types followed consensus frameworks: PJI by International Consensus Meetings (ICM)/Musculoskeletal Infection Society (MSIS)^28,29; FRI by the FRI Consensus Group³⁰; soft-tissue and septic arthritis by Centers for Disease Control and Prevention (CDC)/Infectious Diseases Society of America (IDSA) and StatPearls.^31,32 The European Bone and Joint Infection Society (EBJIS) definition of PJI was also referenced during harmonization.³³ We defined phase as acute (≤6 weeks), chronic (>6 weeks), or acute-on-chronic based on consensus/diagnostic guidance.^33,34 Anatomy was grouped as bone, joint, or soft tissue. Isolates were curated into 11 clinically relevant groups with “negative culture” as reference; organism count was monomicrobial, dual, or polymicrobial (≥3). Low-frequency/collinear organism/comorbidity categories were collapsed (e.g., Gram class) to satisfy events-per-variable (EPV) and variance-inflation (VIF) constraints. Because patients could have >1 organism or comorbidity, these were coded as binary dummies (presence/absence). Complications were classified by Clavien–Dindo (grades ≥III as major).³⁵ Comorbidities were standardised into a few clinically relevant ones (diabetes, chronic kidney disease, end-stage renal disease, rheumatoid arthritis, other autoimmune conditions), with composites (renal impairment; autoimmune-any). For durability, we constructed an interval-censored dataset with left/right bounds from event dates (failure), last observation, and follow-up.

Statistical analysis

Descriptive statistics and hypothesis testing

Continuous summary data (observed only, non-imputed) are presented as medians with interquartile ranges (IQR); categorical summaries use counts (n/N) and percentages (%).

Categorical comparisons employed Pearson’s χ² when expected counts were ≥5 and simulated Fisher’s exact testing for sparser data. One-vs-all contrasts (each subgroup vs all remaining patients) were analysed via two-sample Z-tests where both margins exceeded 30. Adjustments for multiple comparisons were made using Bonferroni and false discovery rate (FDR) corrections. These analyses were repeated stratified by infection type and endpoint, with missing data excluded from hypothesis tests but tallied separately for transparency. Hypothesis tests were exploratory/descriptive given the case report/series design and heterogeneous ascertainment.

We additionally extracted study-level CLAP implementation variables to address heterogeneity in clinical application, including CLAP variant (iMAP/iSAP/iJAP), perfusate antibiotic(s), reported concentration and dose, infusion rate, and intended duration. Where reported, we also captured systemic antibiotic co-therapy (presence, agents, duration) and serum antibiotic level monitoring (maximum reported blood concentration). These protocol features were summarized descriptively and were not treated as analytic targets. Categorical variables are presented as counts/percentages (study-level; each study counted once), and numeric parameters are summarized as medians with interquartile ranges where available and comparable across reports. Full study-level CLAP implementation details are provided in the “Supplementary CLAP Implementation Table”.

Multiple imputation for missing data handling

Missing data were handled with multiple imputation by chained equations (MICE) with m = 20 imputations to reduce Monte-Carlo error. The same imputations were used for descriptive Kaplan–Meier/RMST displays and for the primary regression analysis. Categorical predictors used logistic or polytomous logistic imputation; continuous predictors used predictive mean matching; estimates and standard errors were pooled using Rubin’s rules. Time-to-response event times were not imputed for time-to-event analyses; when a clearance/response time was not reported, observations were right-censored at the last observed follow-up time. The 30-days endpoint was computed directly from reported information (response time when available and follow-up/status otherwise) and was not re-derived from imputed event times. Robustness checks (Supplemental Appendix) included leave-one-study-out analyses for the primary model, exclusion of very short follow-up (<90 days), and the strict durability definition noted above.

Kaplan–Meier and restricted mean survival time (RMST)

For time-to-response displays, an event was defined as a reported time to infection clearance/response. If a clearance time was not reported, the observation was treated as right-censored at the last observed follow-up time. Cases described only as “cleared by 30 days” without a clearance time contributed to the 30-days binary endpoint but were not assigned an event time in time-to-event analyses. Time to treatment response was summarised using Kaplan–Meier curves, pooled across multiply imputed datasets using a complementary log–log transformation with 95% confidence intervals, and RMST at 30 days to characterise trajectory.

For subgroup comparisons, we additionally reported RMST differences (ΔRMST) up to τ = 30 days, estimated within each imputed dataset. For displayed KM subgroup contrasts, log-rank tests, single-variable Cox model Wald tests, and Schoenfeld-residual proportional hazards tests were computed within each imputed dataset and combined across imputations. For figure readability, curves were truncated at 365 days (or earlier if maximum observed follow-up was shorter), while inferential tests were computed using the full available follow-up; censoring was indicated by tick marks. Additional details are shown in the Supplemental Appendix.

Regression analyses

For the primary analysis, we fitted a one-stage, multivariable mixed-effects logistic regression with a random intercept for study (patients nested within studies) across the multiply imputed IPD (m = 20), pooling estimates by Rubin’s rules. The IPD comprised individual patients extracted from multiple primary studies (case reports/series). Each patient record was linked to a study identifier (study_id), and patients were therefore nested within studies. In the regression dataset, after restricting to patients with covariate availability for the prespecified model set, 67 studies contributed patients; cluster size ranged from 1 to 15 patients per study (median 1, IQR 1–2), and ∼61% of studies contributed a single patient. Patients from the same study share study-level features (e.g., CLAP protocol variant, antibiotic choice/dosing, surgical approach, outcome ascertainment, and follow-up intensity), which can induce within-study correlation and between-study heterogeneity. We therefore modelled study as a clustering level using a random intercept for study.

Prespecified covariates were implant involvement, bacteria class (Gram-positive, Gram-negative, anaerobes/fungi/miscellaneous, and culture-negative), mono-versus polymicrobial status (≥2 organisms), infection site/phase, age group, and comorbidity indicators (diabetes, renal impairment, and autoimmune conditions). We report adjusted odds ratios (aORs) with 95% confidence intervals; p-values are descriptive. As robustness checks (Table S7), we fitted a pooled fixed-effects logistic model and a Firth bias-reduced logistic model using the same covariates to address potential separation.

Durable infection control

For durable infection control, we estimated success proportions at 180 and ≥365 days (approximately ≥6 and ≥12 months) using evaluable denominators with exact confidence intervals, and we presented best–worst bounds by allocating unknowns entirely to success or failure. Failure (durability event) was defined as recurrence/relapse, re-operation for infection, or infection-related death. Interval-censored endpoints were constructed as: success (no failure) → (follow-up, ∞), known-time failure → [t, t], and failures without a known event time but with follow-up → (0, follow-up). We also estimated IFD using the non-parametric maximum-likelihood estimator for interval-censored data (Turnbull NPMLE) and integrated the survival curve to τ = 180 and 365 days to obtain RMST.³⁶ Computation used icenReg.³⁷ RMST provides an interpretable summary of average infection-free time and is robust when proportional hazards do not hold.^38,39 Uncertainty was quantified with cluster bootstrap resampling at the study level (studies resampled with replacement, including all patients within each selected study; details in the Supplement).

Results

Study selection and IPD obtained

83 studies (including one District General Hospital series contributing eight additional patients) met inclusion. We requested IPD from all corresponding authors. Where author-provided IPD was unavailable, we reconstructed patient-level data from published tables/figures/narratives when feasible. Two studies lacked usable patient-level detail and one were excluded (another study had one salvageable representative case); one additional study was excluded for duplicate IPD. No study was excluded solely because author-provided IPD was unavailable if sufficient patient-level data could be reconstructed. Across the remaining 81 studies, published/provided IPD yielded 256 patients. For descriptive time summaries, 164 patients had observed time-to-early-response; model-based analyses used multiply imputed datasets. The PRISMA-IPD flowchart is shown in Figure 2.

Figure 2.

PRISMA-IPD flow diagram of study screening, eligibility, IPD retrieval, and inclusion. Abbreviations: IPD: individual participant data; DGH: district general hospital.

Study and patient characteristics

The included observational reports were predominantly small case reports/series with largely similar but slightly heterogenous CLAP protocols. Pooled follow-up was substantial (median 16.5 months, IQR 10.0–27.0). Among the 164 evaluable patients (observed only, non-imputed), median age was 70.0 years (IQR 58.0–77.2), and 53.7% were male. Case mix: PJI 33.5%, FRI 23.2%, osteomyelitis 19.5%, septic arthritis 4.9%, soft-tissue infection 11.0%, and other implant-associated infection 6.7%. Implant involvement was present in 55.8% (10 missing). Lower-limb sites predominated (71.3%). Staphylococcus aureus (MSSA/MRSA) was most frequent (MRSA 22.2%, MSSA 18.8%); Enterococcus spp. 5.6%. Polymicrobial (≥3) infections occurred in 5.5%. Median observed time-to-early-response was 26.0 days (IQR 14.0–55.8). (Table 2; Figure 3(a)–(j); Table S2; Table S3). Table 2 reports row percentages (the proportion with 30-days early response within each subgroup).

Table 2.

Baseline characteristics (observed only) by treatment response within 30 days.

Variable	Overall (n = 164)^a	Early response (+) (n = 97)	Early response (−) (n = 67)	P value	Test used	Bonferroni	FDR
Median time for treatment response [days, IQR]	26.0 [14.0–55.8]	14.0 [10.0–19.0]	56.0 [42.0–86.5]	<0.001	Mann-Whitney U test	NA	NA
Age
Median age at diagnosis [year, IQR]	70.0 [58.0–77.2]	69.0 [57.0–77.0]	71.0 [59.0–77.5]	0.558	Mann-Whitney U test	NA	NA
<65	54 (32.9%)	32 (59.3%)	22 (40.7%)	0.984	Z-test	1.000	0.984
≥65	110 (67.1%)	65 (59.1%)	45 (40.9%)
Number of patients with missing data	0	0	0
Sex
Male	88 (53.7%)	52 (59.1%)	36 (40.9%)	0.988	Z-test	1.000	0.988
Female	76 (46.3%)	45 (59.2%)	31 (40.8%)
Number of patients with missing data	0	0	0
Co-morbidities/Underlying conditions^b
Diabetes mellitus	52 (31.7%)	29 (55.8%)	23 (44.2%)	0.762	Z-test	1.000	1.000
Hypertension	24 (14.6%)	11 (45.8%)	13 (54.2%)	0.367	Chi-squared	1.000	0.979
Dyslipidemia	10 (6.1%)	5 (50.0%)	5 (50.0%)	1.000	Fisher	1.000	1.000
Rheumatoid arthritis	13 (7.9%)	5 (38.5%)	8 (61.5%)	0.232	Chi-squared	1.000	0.948
Other autoimmune conditions	5 (3.0%)	2 (40.0%)	3 (60.0%)	0.660	Fisher	1.000	1.000
None	35 (21.3%)	18 (51.4%)	17 (48.6%)	0.708	Z-test	1.000	1.000
Number of patients with missing data	31	25	6
Infection type
Periprosthetic joint infection	55 (33.5%)	25 (45.5%)	30 (54.5%)	0.011	Z-test	0.079	0.039
Fracture-related infection	38 (23.2%)	22 (57.9%)	16 (42.1%)	0.858	Z-test	1.000	1.000
Osteomyelitis	32 (19.5%)	28 (87.5%)	4 (12.5%)	<0.001	Z-test	0.002	0.002
Soft tissue infection	18 (11.0%)	12 (66.7%)	6 (33.3%)	0.492	Chi-squared	1.000	0.719
Other implant-associated infection	11 (6.7%)	3 (27.3%)	8 (72.7%)	0.052	Fisher	0.363	0.121
Septic arthritis	8 (4.9%)	5 (62.5%)	3 (37.5%)	1.000	Fisher	1.000	1.000
Necrotizing soft tissue infection	2 (1.2%)	2 (100.0%)	0 (0.0%)	0.514	Fisher	1.000	0.719
Number of patients with missing data	0	0	0
Implant involvement
Yes	86 (55.8%)	37 (43.0%)	49 (57.0%)	<0.001	Z-test	0.000	0.000
No	68 (44.2%)	50 (73.5%)	18 (26.5%)
Number of patients with missing data	10	10	0
Infection location^c (overall)
Upper limb	34 (20.7%)	26 (76.5%)	8 (23.5%)	0.021	Z-test	0.084	0.084
Lower limb	117 (71.3%)	65 (55.6%)	52 (44.4%)	0.140	Z-test	0.560	0.280
Spine	12 (7.3%)	5 (41.7%)	7 (58.3%)	0.232	Fisher	0.928	0.309
Others	1 (0.6%)	1 (100.0%)	0 (0.0%)	1.000	Fisher	1.000	1.000
Number of patients with missing data	0	0	0
Anatomic site by group
Bone	81 (49.4%)	53 (65.4%)	28 (34.6%)	0.106	Z-test	0.317	0.159
Joint	63 (38.4%)	30 (47.6%)	33 (52.4%)	0.018	Z-test	0.053	0.053
Soft tissue	20 (12.2%)	14 (70.0%)	6 (30.0%)	0.292	Chi-squared	0.876	0.292
Number of patients with missing data	0	0	0
Infection phase
Acute	59 (41.0%)	31 (52.5%)	28 (47.5%)	0.744	Z-test	1.000	1.000
Chronic	83 (57.6%)	46 (55.4%)	37 (44.6%)	0.724	Z-test	1.000	1.000
Acute on chronic	2 (1.4%)	1 (50.0%)	1 (50.0%)	1.000	Fisher	1.000	1.000
Number of patients with missing data	20	19	1
Microbial type
Other Staphylococcus spp. & CoNS	34 (23.6%)	17 (50.0%)	17 (50.0%)	0.542	Z-test	1.000	0.842
MRSA	32 (22.2%)	14 (43.8%)	18 (56.2%)	0.164	Z-test	1.000	0.601
MSSA	27 (18.8%)	16 (59.3%)	11 (40.7%)	0.585	Chi-squared	1.000	0.842
Anaerobes, fungi & miscellaneous	20 (13.9%)	7 (35.0%)	13 (65.0%)	0.058	Chi-squared	0.642	0.321
Enterobacteriaceae	16 (11.1%)	7 (43.8%)	9 (56.2%)	0.357	Chi-squared	1.000	0.786
ESBL producing organism	5 (3.5%)	2 (40.0%)	3 (60.0%)	0.659	Fisher	1.000	0.842
Enterococcus spp.	8 (5.6%)	0 (0.0%)	8 (100.0%)	0.001	Fisher	0.016	0.016
Negative culture	9 (6.2%)	5 (55.6%)	4 (44.4%)	1.000	Fisher	1.000	1.000
Other gram-negative bacilli	6 (4.2%)	4 (66.7%)	2 (33.3%)	0.689	Fisher	1.000	0.842
Other Streptococcus spp.	10 (6.9%)	7 (70.0%)	3 (30.0%)	0.347	Fisher	1.000	0.786
Pseudomonas aeruginosa	12 (8.3%)	7 (58.3%)	5 (41.7%)	0.783	Chi-squared	1.000	0.861
Number of patients with missing data	21	19	2
Number of organisms detected
Number of organisms [median, IQR]	1.0 [1.0–1.0]	1.0 [1.0–1.0]	1.0 [1.0–1.5]	0.013	Mann–Whitney U	NA	NA
Monomicrobial	135 (82.8%)	85 (63.0%)	50 (37.0%)	0.020	Chi-squared	0.061	0.031
2 organisms	19 (11.7%)	10 (52.6%)	9 (47.4%)	0.555	Chi-squared	1.000	0.555
Polymicrobial (≥3)	9 (5.5%)	1 (11.1%)	8 (88.9%)	0.004	Fisher	0.011	0.011
Number of patients with missing data	1	1	0

^aOf 256 total patients, 164 had evaluable 30-days early-response status (CRP ≤3 mg/L or clinical/wound improvement within 30 days) and were included in Table 2. The remaining 92 were unevaluable for 30-days response due to missing outcome documentation (and/or insufficient follow-up where available).

^bOnly clinically relevant comorbidities are shown. Full list of comorbidities are shown in the Supplementary Appendix (Table S2).

^cFor brevity, detailed “Infection Location (Anatomic)” tables are not included in this table but provided in the Supplementary Appendix (Table S2).

IQR = interquartile range; FDR = false discovery rate; NA = not applicable; MRSA = Methicillin-resistant Staphylococcus aureus; MSSA = Methicillin-susceptible Staphylococcus aureus; ESBL = Extended-spectrum beta-lactamase; CoNS = coagulase-negative Staphylococci.

Figure 3.

Pooled summary mosaic plot of demographics, comorbidities, infection characteristics, microbiology, outcomes, and follow-up in CLAP-treated patients; Patient IDs: (A) 1A to 5J, (B) 6 to 12N, (C) 13A to 15J, (D) 16 to 24H, (E) 25 to 44, (F) 45A to 53N, (G) 53.1A to 57B, (H) 58A to 69, (I) 70A to 73C, (J) 74A to 82. Refer to (J) for color and abbreviation legends.

Implementation of CLAP across studies

Across the 81 included studies, CLAP implementation details were commonly reported but heterogeneous (see Supplemental CLAP Implementation Table). CLAP configuration (iMAP/iSAP/iJAP) was reported in 75/81 studies (92.6%), and perfusate antibiotic choice in 76/81 (93.8%) (Table 1). Gentamicin was the predominant perfusate antibiotic (67/81; 82.7%). Where infusion parameters were available, reported regimens clustered around a common operating profile, with a median concentration of 1.20 mg/mL (64/81 studies reported available concentration), infused at 2.00 mL/h (63/81 available), for a median duration of 14.00 days (74/81 available). Systemic antibiotic co-therapy was frequently reported (65/81; 80.2%), with median reported duration of 28.00 days among studies reporting available duration (54/81)

Table 1.

Summary of CLAP protocol implementation across included studies (study-level, n = 81).

Reporting completeness (study-level, n = 81)	n/N (%)	Typical CLAP operating parameters (study-level)	Median [IQR] (n parsable)	CLAP configuration (iMAP/iSAP/iJAP)	n/N (%)	Perfusate antibiotic	n/N (%)	Systemic antibiotics	n/N (%)
CLAP type reported	75/81 (92.6%)	Concentration (mg/mL)	1.20 [1.20, 1.20] (n = 64)	iMAP + iSAP	23/81 (28.4%)	Gentamicin	67/81 (82.7%)	IV antibiotics reported as used (yes)	65/81 (80.2%)
Perfusate antibiotic reported	76/81 (93.8%)	Rate (mL/h)	2.00 [2.00, 2.00] (n = 63)	iSAP	22/81 (27.2%)	Amikacin	2/81 (2.5%)	IV antibiotics not reported	16/81 (19.8%)
Concentration parsable (mg/mL)	64/81 (79.0%)	CLAP duration (days)	14.00 [14.00, 19.88] (n = 74)	iMAP	16/81 (19.8%)	Other perfusates (combined)	7/81 (8.6%)
Rate parsable (mL/h)	63/81 (77.8%)	IV duration (days)	28.00 [14.00, 42.00] (n = 54)	iJAP	8/81 (9.9%)	Not reported	5/81 (6.2%)
CLAP duration parsable (days)	74/81 (91.4%)			Other configurations (combined)	6/81 (7.4%)
IV antibiotic use reported	65/81 (80.2%)			Not reported	6/81 (7.4%)
IV duration parsable (days)	54/81 (66.7%)

Values are study-level summaries; each study contributes once. Parsed fields standardize concentration (mg/mL), rate (mL/h), and durations (days) when extractable. iMAP = intramedullary antibiotic perfusion; iSAP = intra soft-tissue antibiotic perfusion; iJAP = intra-articular antibiotic perfusion.

Primary outcome—30-days early response

Among 245 patients evaluable for the 30-days binary endpoint, the MI-pooled 30-days response proportion was 67.0% (95% CI 61.0–72.5) (Table 2; Table 3). Among the 164 patients with observed time-to-response, 97 (59.1%) achieved early response by 30 days and 67 (40.9%) did not. Median time-to-response was 26.0 days (IQR 14.0–55.8); 14.0 days (IQR 10.0–19.0) among successes versus 56.0 days (IQR 42.0–86.5) among non-successes. Infection type was associated with early success (global p ≤ 0.001): with higher response in osteomyelitis (87.5%) and septic arthritis (62.5%), and lower response in PJI (45.5%) and other implant-associated infection (27.3%). Implant involvement was associated with reduced success (43.0% with implants vs 73.5% without). Upper-limb infections were more often successful (76.5%); joint site had lower success than bone or soft tissue (47.6%). Increasing organism burden was also associated with lower observed early response (monomicrobial 63.0%, two organisms 52.6%, polymicrobial ≥3 11.1%). Enterococcus spp. showed particularly poor observed early response (0% in the observed data). (Table 2).

Table 3.

Mixed-effects multivariable logistic analysis (random study intercept) for predictors of 30-days early treatment response.

Variable	Adjusted odds ratio (aOR)	95% CI	P value
Age
<65	[Ref]
≥65	2.59	(0.72–9.38)	0.147
Co-morbidities/Underlying conditions
None	[Ref]
Diabetes mellitus	0.95	(0.31–2.94)	0.925
Renal impairment^a	2.15	(0.18–26.11)	0.548
Autoimmune conditions^b	2.37	(0.33–16.97)	0.389
Pathoanatomic site
Bone	[Ref]
Joint	0.55	(0.08–3.97)	0.557
Soft tissue	0.28	(0.03–2.30)	0.235
Implant involvement
No	[Ref]
Yes	0.18	(0.04–0.79)	0.023
Infection location
Lower limb	[Ref]
Upper limb	0.8	(0.12–5.44)	0.822
Infection phase
Acute	[Ref]
Acute on chronic	0.16	(0.00–16.12)	0.437
Chronic	1.06	(0.30–3.71)	0.924
Microbial type
Negative culture	[Ref]
Gram positive bacteria	1.64	(0.11–24.75)	0.721
Gram negative bacteria	1.31	(0.07–25.14)	0.857
Anaerobes, fungi and miscellaneous	1.72	(0.04–79.08)	0.78
Number of organisms detected
Monomicrobial	[Ref]
Polymicrobial (≥2 organisms)^c	0.32	(0.06–1.79)	0.193

To maintain adequate events-per-variable (EPV) and model stability, predictors were collapsed/remapped; infection type and sparse/collinear variables were excluded. Specifications are detailed in the Supplementary Appendix.

CI = confidence interval; Ref = reference group.

^aComposite of chronic kidney disease & end stage renal failure on dialysis.

^bComposite of rheumatoid arthritis & other autoimmune diseases.

^cComposited of “2 organisms” and “polymicrobial (≥3)”.

Kaplan–Meier—Cumulative probability of treatment response

In Kaplan–Meier analysis, the MI-pooled estimated 30-days cumulative probability of treatment response was 45.6% (95% CI 38.9–53.0), which is notably lower compared to the non-MI-pooled subset (13.5% difference). This KM-based estimate is lower than the binary 30-days response proportion because cases described only as ‘cleared by 30 days’ contribute to the binary endpoint but are not assigned an event time and are treated as censored in time-to-event analyses.

RMST up to 30 days was 23.41 days (95% CI 22.19–24.63) (Figure 4(a)–(d)). Across subgroup curves, statistically significant differences in time to response were observed by age group (pooled log-rank p ≤ 0.001), sex (p ≤ 0.001), pathoanatomic site (p ≤ 0.001), anatomical region (p ≤ 0.001), and infection phase (p ≤ 0.001). Cumulative response increased rapidly early after treatment initiation and then plateaued, with progressively fewer patients remaining at risk later in follow-up.

Figure 4.

Kaplan–Meier curves for time to treatment response stratified by: (A) age; (B) sex; (C) pathoanatomic site; (D) implant involvement; (E) anatomical region; (F) infection phase; (G) microbial class; (H) organism count. Shaded bands, 95% CI; Curves are truncated at 365 days for display (max follow-up 1826 days); the red dotted line marks the 30-day cutoff. Tick marks indicate censoring; numbers at risk are shown.

Regression analysis

In the one-stage mixed-effects multivariable logistic model (random intercept for study), implant involvement remained independently associated with lower odds of 30-days early response (aOR 0.18, 95% CI 0.04–0.79; p = 0.023). Other candidate predictors (age group, sex, infection type contrasts, and polymicrobial infection) were imprecise and were not statistically significant after adjustment. (Table 3).

RMST for early-response trajectory (τ = 30 days)

Cluster-adjusted RMST contrasts (τ = 30 days) indicated slower early-response trajectories with implant involvement (Yes vs No: ΔRMST +4.43 days, 95% CI 0.89–7.73; p = 0.016) and polymicrobial infection (≥3) compared with monomicrobial infection (ΔRMST +6.74 days, 95% CI 4.28–9.59; p ≤ 0.001). Among organism subgroups, Enterococcus spp. was associated with a markedly slower trajectory (ΔRMST +6.93 days, 95% CI 4.63–9.56; p ≤ 0.001), and the anaerobes/fungi/miscellaneous group also showed slower trajectories (ΔRMST +3.99 days, 95% CI 0.58–6.97; p = 0.025). In contrast, osteomyelitis demonstrated a faster trajectory relative to FRI (ΔRMST −9.06 days, 95% CI −12.87 to −4.22; p ≤ 0.001). Other contrasts, including two organisms versus monomicrobial (ΔRMST +0.75 days, 95% CI −3.36–4.81; p = 0.719), were not statistically significant. (Table 4).

Table 4.

Cluster-bootstrap adjusted restricted mean time to treatment response differences (ΔRMST, τ = 30 days). Positive values indicate delay to clearance.

Variable	τ (Days)	ΔRMST	95% CI	P value
Age
≥65 vs. <65	30	−0.17	(−3.39–2.94)	0.895
Sex
Female vs. Male	30	−0.79	(−3.49–1.70)	0.534
Co-morbidities/Underlying conditions^a (present vs. absent)
None vs. Comorbidities present	30	0.00	(−3.16–3.27)	0.584
Diabetes mellitus	30	0.63	(−2.35–3.58)	0.678
Chronic kidney disease	30	1.65	(−3.54–6.76)	0.500
End stage renal disease on dialysis	30	−1.18	(−8.61–5.91)	0.726
Rheumatoid arthritis	30	1.63	(−7.66–7.69)	0.628
Other autoimmune conditions	30	0.58	(−7.76–7.61)	0.843
Infection type
Other IAI vs. FRI	30	1.63	(−3.98–5.44)	0.440
Osteomyelitis vs. FRI	30	−9.06	(−12.87–-4.22)	<0.001
PJI vs. FRI	30	−2.92	(−9.06–2.16)	0.287
Septic arthritis vs. FRI	30	−5.01	(−12.22–3.72)	0.274
Soft tissue infection vs. FRI	30	−5.51	(−11.34–0.76)	0.088
Necrotizing soft tissue infection vs. FRI	30	−9.56	(−17.70–5.41)	0.540
Pathoanatomic site
Joint vs. Bone	30	−0.44	(−5.83–4.05)	0.878
Soft tissue vs. Bone	30	−3.23	(−8.40–2.33)	0.264
Implant involvement
Yes vs. No	30	4.43	(0.89–7.73)	0.016
Infection location
Upper limb vs. Lower limb	30	−6.06	(−11.09–0.96)	0.1
Spine vs. Lower limb	30	1.68	(−8.78–5.88)	0.609
Others vs. Lower limb	30	3.45	(−10.80–7.23)	0.573
Infection phase
Acute on chronic vs. Acute	30	−2.55	(−18.54–7.05)	0.815
Chronic vs. Acute	30	−2.76	(−6.78–1.43)	0.222
Microbial type (present vs. absent)
Negative culture vs. Bacteria detected	30	0.45	(−6.57–6.74)	0.829
Anaerobes, fungi & miscellaneous	30	3.99	(0.58–6.97)	0.025
ESBL producing organism	30	0.20	(−8.98–7.69)	0.940
Enterobacteriaceae	30	1.07	(−4.27–4.66)	0.643
Enterococcus spp.	30	6.93	(4.63–9.56)	<0.001
MRSA	30	1.41	(−1.92–4.38)	0.391
MSSA	30	0.94	(−2.90–3.56)	0.612
Other gram-negative bacilli	30	2.49	(−4.01–7.16)	0.394
Other Staphylococcus spp. & CoNS	30	−0.30	(−4.06–3.42)	0.832
Other Streptococcus spp.	30	−5.06	(−11.21–1.11)	0.106
Pseudomonas aeruginosa	30	−2.57	(−10.50–5.62)	0.621
Number of organisms detected
2 organisms vs. Monomicrobial	30	0.75	(−3.36–4.81)	0.719
Polymicrobial (≥3) vs. Monomicrobial	30	6.74	(4.28–9.59)	<0.001

CI = confidence interval; FRI = Fracture-Related Infection; IAI = Implant-Associated Infection; PJI = Periprosthetic Joint Infection; MRSA = Methicillin-resistant Staphylococcus aureus; MSSA = Methicillin-susceptible Staphylococcus aureus; ESBL = Extended-spectrum beta-lactamase; CoNS = Coagulase-negative Staphylococci.

^aOnly clinically relevant comorbidities are shown.

Secondary outcome—Durable infection control & infection-free days

Durable control at prespecified windows (Table 5) used evaluable denominators (patients with follow-up at/after the window and determinate status), with transparent best–worst bounds for unknowns. At 6 months (τ = 180 days): FU ≥ τ 157, evaluable 146, success 88.4% (95% CI 82.0–93.1), unknown = 11; best–worst bounds 89.2–82.2%. At 12 months (τ = 365 days): FU ≥ τ 121, evaluable 112, success 90.2% (95% CI 83.1–95.0), unknown = 9; bounds 90.9–83.5%. Infection-Free Days (Turnbull RMST) were 170.5 days (95% CI 163.6–180.0) to 180 days and 333.9 days (95% CI 315.5–355.6) to 365 days. Subgroup contrasts (Tables S10-S11) were clinically coherent with the primary endpoint (unfavourable with implant involvement and higher organism burden; favourable for osteomyelitis).

Table 5.

Durable infection control (windows) and infection-free days (Turnbull RMST).

Measure	Horizon (τ)	FU ≥ τ (N)	Evaluable (n)	Durability, % (95% CI)	Unknown (n)	Best–worst bounds, %
Long term infection control	180 days	157	146	88.4 (82.0–93.1)	11	89.2–82.2
Long term infection control	365 days	121	112	90.2 (83.1–95.0)	9	90.9–83.5
RMST (infection-free days)	180 days	—	—	170.5 days (163.6–180.0)	—	—
RMST (infection-free days)	365 days	—	—	333.9 days (315.5–355.6)	—	—

CI = confidence interval; FU = follow-up; RMST = restricted mean survival time.

Evaluable n are cases with sufficient follow-up to determine status at τ; unknown n lack sufficient follow-up, so best/worst success bounds treat unknowns as all success/all failure. Turnbull interval-censored RMST estimates mean infection-free days up to τ (days and % of τ) with clustered bootstrap 95% CIs.

Safety

Adverse events were infrequent, occurring in 13/175 patients (7.43%): four major (2.29%) and nine minor (5.14%). Where reported, renal adverse events were generally reversible with dose adjustment or cessation, and no ototoxic events were described. Major complications (Clavien–Dindo ≥ III) were uncommon. Reporting was heterogeneous across studies; detailed counts and denominators are provided in Table S6 (Supplement).

Discussion

What these analyses tell us—and do not tell us

These results synthesise observational reports and describe early trajectories and durability patterns within the limits of the available data; they are hypothesis-generating rather than causal treatment effects. Although we extracted perfusion parameters where reported, they were inconsistently documented and strongly confounded by indication/anatomy and local program design; therefore, modelling them as dose–response predictors risked spurious inference and we summarised them descriptively.

Summary of results

In this IPD synthesis of 256 CLAP-treated patients (n = 164 evaluable for the 30-days endpoint), 97 (59.1%) achieved early response within 30 days. In the one-stage mixed-effects logistic model (random intercept for study), retained implant involvement was associated with lower odds of 30-days response (aOR 0.18, 95% CI 0.04–0.79). Time-scale summaries using RMST to 30 days were directionally consistent, showing slower trajectories with retained implants (ΔRMST +4.43 days) and polymicrobial ≥3 (ΔRMST +6.74 days), and faster trajectories for osteomyelitis versus FRI (ΔRMST −9.06 days). Durability among evaluable patients remained high at 6 and 12 months: at 180 days, success was 88.4% (129/146) with worst-case bounds 82.2%, and at 365 days, success was 90.2% (101/112) with worst-case bounds 83.5%. Infection-free-day RMST indicated substantial time spent infection-free within the first year (170.5 days to 180 days; 333.9 days to 365 days). Early CRP response does not guarantee eradication, but these durability summaries suggest that for many patients it marks a favourable early trajectory rather than a definitive cure. Major adverse events were uncommon; renal events were uncommon and generally reversible where reported. Overall, CLAP appears may be most useful in contexts where anatomy and microbiology are favourable and less so when hardware is retained, organism burden is high, or difficult organisms are present.^{3,4,7,8,10–13,40}

Clinical and research implications

For patients with retained implants or polymicrobial infection, anticipate a slower early response and plan for ∼4–7 additional days without response (RMST) within the 30-days window. Early escalation of monitoring is appropriate, including renal function and drug levels when aminoglycosides are used, and source control should be pursued without jeopardizing essential hardware. Where anatomy and microbiology are favourable—for example shoulder infection with C. acnes—CLAP with a disciplined ingress–egress circuit may accelerate early improvement and support discharge planning when paired with effective debridement and systemic antibiotics.^41–43 Programs adopting CLAP should standardize catheter maps by infection pattern, provide drug-preparation guides with concentration ranges and flow or aspiration targets, use securement and monitoring checklists, and establish laboratory schedules, especially for older adults or longer courses. Embedding these elements in order sets and maintaining a minimal prospective dataset that tracks time to early response, recurrence, complications, unplanned returns to theatre, and simple function flags will support continuous quality improvement and value analyses. Prospective registries or trials should standardize CLAP protocols, collect pharmacokinetics, and power organism-level contrasts, with pre-specified RMST endpoints to accommodate non-proportional hazards and harmonized CRP-based definitions to improve comparability.^{15–17,19,36–39,44}

Implementation heterogeneity and a common operating profile

Across included studies, CLAP implementation was variably reported but clustered around a recognizable operating profile: configuration was reported in 92.6% of studies, perfusate choice in 93.8%, with gentamicin predominating (82.7%). Where available, regimens commonly reported a concentration of ∼1.20 mg/mL infused at ∼2.0 mL/h for ∼14 days, and most studies reported concurrent systemic antibiotics (80.2%) with a median reported duration of ∼28 days. Although the cohort remains clinically heterogeneous across indications, anatomy, and institutional contexts, we observed a recognizable convergence in CLAP implementation across reports. This relative consistency of the intervention exposure improves the interpretability of pooled descriptive estimates by reducing heterogeneity attributable to protocol variation. These summaries help contextualize external validity and stewardship, but protocol fields were not consistently extractable and were not modelled as dose–response predictors because reporting was inconsistent and confounded by indication and program design, so implementation–outcome relationships remain uncertain and are best addressed in prospective registries/trials.

Biology and burden of complexity

The penalties associated with retained implants and polymicrobial communities are biologically plausible. Foreign material supports mature biofilm; mixed consortia produce quorum-mediated protection and metabolic layering that raise eradication thresholds above planktonic MICs.^3,4,10,40 CLAP is designed to confront these barriers by sustaining high local concentration–time exposure at the nidus and evacuating inhibitory exudate through controlled egress (typically via NPWT).^7,8,11–13 Debridement remains foundational; CLAP appears to preserve that surgical gain by bathing poorly perfused interfaces where systemic therapy struggles to deliver adequate exposure. The ∼4–9-days RMST differences we observed are modest in absolute terms yet clinically meaningful in the early window when escalation, re-operation, or discharge decisions are made.

Infection type and anatomy

Osteomyelitis improved comparatively quickly, plausibly because intramedullary or perilesional lines access canal and dead-space surfaces for sustained exposure.^7,8 Given osteomyelitis can harbour indolent infection, a 30-days response should be interpreted cautiously; however, our 6- and 12-months follow-up suggests that many early responders did achieve sustained control. Periprosthetic joint infection was more refractory, echoing multicentre cohorts that underscore the difficulty of eradicating biofilm on prosthetic surfaces and the importance of decisive source control.²¹ Recent chronic knee PJI case series also report feasibility of CLAP during DAIR.⁴⁵ Early successes in the shoulder and upper limb may reflect organism spectrum (including Cutibacterium acnes), thinner soft-tissue envelopes, and the feasibility of intra-articular perfusion with active aspiration.^41–43,46 In contrast, large lower-limb constructs and cement mantles create diffusion barriers and potential micro-reservoirs that slow early response.

Safety, stewardship, and monitoring

Adverse events were infrequent, but nephrotoxicity warrants vigilance. An independent cohort linked early renal decline to older age and longer CLAP duration, supporting conservative course lengths, good hydration, avoidance of nephrotoxins, and routine creatinine monitoring in older adults.⁴⁷ Drug levels can be considered for aminoglycosides when infusions extend beyond about a week or renal reserve is limited, with predefined rules for adjustment or cessation. Routine screening for otologic and vestibular symptoms is prudent. Concerns that local antibiotics drive resistance are not supported by the best available comparative evidence in extremity fracture-related infection, although stewardship still applies: narrow spectrum once cultures finalize, avoid unnecessary dual coverage, and culture any recurrences to surveil for resistance.⁴⁸

Positioning against established modalities

Selected contemporary series/trials report two-stage exchange for chronic hip and knee PJI can reach about 95% eradication in selected series, and one-stage programs report approximately 90–95% mid-term control.^49–51 On earlier horizons, optimized very-early DAIR within 1 week can achieve about 93% control, and repeated DAIR can salvage 78–83% of selected failures.^52,53 For systemic-therapy context, OVIVA showed oral therapy was non-inferior to intravenous therapy with roughly 85–87% 1-year success.⁵⁴ In osteomyelitis, PMMA beads after thorough debridement report about 91% cure; evidence for calcium-sulphate beads is mixed, with several matched or comparative analyses showing no consistent improvement over DAIR alone and raising cost or complication considerations, including occasional bead- or cement-related reactions.^55–59 Against our 30-days profile, CLAP appears competitive when anatomy and microbiology are favourable (e.g., osteomyelitis and the shoulder) but less attractive when implants are retained or pathogens are multiple. Exchange strategies target longer-horizon eradication that CLAP does not attempt to replace.

Relation to endpoint choice and statistical approach

Focusing on a 30-days early-response endpoint is clinically congruent with early DAIR windows, surveillance timeframes, and quality metrics, and it avoids conflating short-term trajectories with longer-term eradication.^19–23 The Kaplan–Meier curves showed frequent crossings and non-proportional hazards; RMST was therefore used to summarize time-scale differences in a way that remains interpretable when hazards are not proportional.^38,39 For durability, interval-censored NPMLE with explicit evaluable denominators and best–worst bounds kept inference transparent when status was unknown, and infection-free-day RMST captured the lived patient experience of time spent free of recurrent infection.^36,37 These complementary analyses situate CLAP’s role: a technique that can accelerate early improvement without claiming parity with exchange strategies on late eradication.

Strengths

This appears to be the first CLAP IPD synthesis at scale, retrieving and harmonizing patient-level data across 81 studies and applying unified diagnostic frameworks from ICM/MSIS, EBJIS, and the FRI Consensus with standardized outcome timing and definitions.^28–34 The primary endpoint was analysed in a one-stage mixed-effects framework that accounted for study-level clustering. We paired a clinically interpretable 30-days endpoint with interval-censored durability and infection-free-day RMST, methods that are robust to non-proportional hazards and intuitive for clinicians.^36–39,44 Denominators for durability were explicit, and best–worst bounds were reported rather than imputed. IPD integrity checks, multiple imputation with re-derived binary indicators to avoid leakage, and study-cluster bootstrap addressed missingness and between-study heterogeneity. Our outcome framework and analytic approach provide a practical template for future prospective studies and registries evaluating CLAP, particularly where non-proportional hazards and incomplete data are expected.

Limitations

This single-arm synthesis supports contextual, hypothesis-generating inference rather than causal or comparative conclusions versus systemic therapy, beads/spacers, or exchange strategies. Because randomized evidence for CLAP is scarce, most source reports were small observational case reports/series with heterogeneous protocols and variable reporting, increasing clinical heterogeneity and the risk of information bias. Publication and selective reporting biases are likely (e.g., preferential reporting of notable successes or atypical cases), and key clinical variables were inconsistently captured across studies. Outcome ascertainment was heterogeneous (timing and thresholds of CRP testing; variability in how “sustained clinical/wound improvement” was documented), and follow-up was variable, raising the possibility of informative missingness/censoring that cannot be fully corrected statistically. Residual confounding is unavoidable, including confounding by indication (e.g., debridement quality, surgical timing, host factors, and implant involvement thresholds). Events per variable were modest, so subgroup estimates—particularly organism-specific contrasts (e.g., Enterococcus sp.)—are imprecise and should be interpreted as hypothesis-generating prognostic associations rather than treatment effects. Missingness in observed time to response required multiple imputation; sensitivity analyses preserved directionality but precision remained limited. Implementation parameters (e.g., catheter configuration, drug/concentration, flow or aspiration targets, and NPWT settings) varied across reports, although several core elements showed convergence; these were therefore summarised descriptively rather than modelled as dose–response predictors. Chronic suppressive antibiotic use was rarely reported in the IPD and was handled descriptively; conclusions about suppression-dependent durability are limited. Adverse-event ascertainment was variably reported, so rare or delayed toxicities may be under-captured. Generalisability may vary across settings, particularly where devices and monitoring differ.⁶⁰

Conclusion

CLAP can accelerate early response in selected patients when combined with sound debridement and systemic antibiotics. Durability among evaluable patients at six and 12 months was high, and safety events were infrequent and largely reversible, although renal monitoring is essential with aminoglycosides, particularly in older adults and longer courses.⁴⁷ Ultimately, because our analysis is based on uncontrolled data, estimates should be interpreted as hypothesis-generating, and definitive guidance will require prospective trials or registries. Our findings can serve as a benchmark and inform the design of those future studies. Future work should standardize protocols, capture pharmacokinetics, and evaluate comparative effectiveness using RMST-friendly endpoints that acknowledge the common absence of proportional hazards in this field.^38,39

Supplemental material

Supplementary Appendix - Continuous local antibiotic perfusion for orthopaedic infections: A systematic review and pooled individual participant data analysis of observational reports

Supplementary CLAP Implementation Table - Continuous local antibiotic perfusion for orthopaedic infections: A systematic review and pooled individual participant data analysis of observational reports

Supplemental materials for Continuous local antibiotic perfusion for orthopaedic infections: A systematic review and pooled individual participant data analysis of observational reports by Patrick Ze-En Ng, Norio Yamamoto, Mari Yamamoto, Ke Wei Hiew, Wei Ching Chong, Jia Shen Goh, Akihiro Saitsu, Glenn Xin-Zhang Lee, and Naoya Inagaki in Journal of Orthopaedic Surgery

Footnotes

Acknowledgements

We thank Jun Watanabe for his assistance with the EMBASE (ProQuest) database literature search; Satoshi Yamaguchi, Nobuyasu Ochiai, Yukichi Zenke, Keisuke Shimbo, Hideharu Nakamura, Hyonmin Choe, Kenji Kosugi, Kohei Iwamoto, Shuhei Ohyama, Noriaki Yokogawa, Yoshiaki Miyake, Naoyuki Horie, Noriyuki Hattori and Hirotaka Oishi for unpublished IPD upon request.

ORCID iDs

Patrick Ze-En Ng

Norio Yamamoto

Mari Yamamoto

Ke Wei Hiew

Wei Ching Cheong

Jia Shen Goh

Akihiro Saitsu

Glenn Xin-Zhang Lee

Naoya Inagaki

Ethical considerations

This synthesis used published and author-provided de-identified individual participant data from previously reported clinical studies. No new human or animal subjects were recruited. Institutional review board (IRB) approval and informed consent were obtained in the original studies as reported by their authors; additional ethics approval for this IPD synthesis was not required. The protocol was registered with PROSPERO (CRD42025635194) and the review followed PRISMA-IPD.

Author contributions

Conception and Design: Patrick Ng Ze En, Norio Yamamoto. Database search and Acquisition of articles: Patrick Ng Ze En, Hiew Ke Wei, Cheong Wei Ching, Norio Yamamoto, Mari Yamamoto. Risk of Bias Assessment: Hiew Ke Wei, Cheong Wei Ching, Goh Jia Shen, Glenn Lee Xin Zhang, Mari Yamamoto, Akihiro Saitsu, Naoya Inagaki. Data Extraction: Patrick Ng Ze En, Hiew Ke Wei, Cheong Wei Ching, Goh Jia Shen, Glenn Lee Xin Zhang, Mari Yamamoto, Akihiro Saitsu, Naoya Inagaki. Acquisition of Unpublished Data: Norio Yamamoto. Statistical Analysis: Patrick Ng Ze En. Manuscript Draft: Patrick Ng Ze En. Critically Revising the Article: Patrick Ng Ze En, Norio Yamamoto, Hiew Ke Wei, Cheong Wei Ching, Goh Jia Shen, Glenn Lee Xin Zhang, Mari Yamamoto, Akihiro Saitsu. Clinical Expertise: Norio Yamamoto, Mari Yamamoto, Akihiro Saitsu.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Datasets were generated and may be available upon request from the corresponding author.*

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the author(s) used ChatGPT 4o and 5 (Open AI) as a tool for data analysis assistance, grammar correction and formatting compliance. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Supplemental material

Supplemental material for this article is available online.

References

Gill

Mittinty

March

, et al. Global, regional, and national burden of other musculoskeletal disorders, 1990–2020, and projections to 2050: a systematic analysis of the global burden of disease study 2021. Lancet Rheumatol 2023; 5(11): e670–e682. https://doi.org/10.1016/S2665-9913(23)00232-1

Metsemakers

W-J

Moriarty

Morgenstern

, et al.

The global burden of fracture-related infection: can we do better?

Lancet Infect Dis 2024; 24(6): e386–e393. https://doi.org/10.1016/S1473-3099(23)00503-0

Morgenstern

Kuehl

Zalavras

, et al. The influence of duration of infection on outcome of debridement and implant retention in fracture-related infection. Bone Jt J 2021; 103-B(2): 213–221. https://doi.org/10.1302/0301-620X.103B2.BJJ-2020-1010.R1

Mathews

Weston

Jones

, et al. Bacterial septic arthritis in adults. Lancet 2010; 375(9717): 846–855. https://doi.org/10.1016/S0140-6736(09)61595-6

Balato

Ascione

de Matteo

, et al. Debridement and implant retention in acute hematogenous periprosthetic joint infection after knee arthroplasty: a systematic review. Orthop Rev (Pavia) 2022; 14(5): 33670. https://doi.org/10.52965/001c.33670

Kunutsor

Beswick

Whitehouse

, et al. Debridement, antibiotics and implant retention for periprosthetic joint infections: a systematic review and meta-analysis of treatment outcomes. J Infect 2018; 77(6): 479–488. https://doi.org/10.1016/j.jinf.2018.08.017

Ford

Cassat

. Advances in the local and targeted delivery of anti-infective agents for management of osteomyelitis. Expert Rev Anti Infect Ther 2017; 15(9): 851–860. https://doi.org/10.1080/14787210.2017.1372192

Maruo

Oda

Mineo

, et al. Continuous local antibiotic perfusion: a treatment strategy that allows implant retention in fracture-related infections. J Orthop Surg 2022; 30(2): 10225536221111902. https://doi.org/10.1177/10225536221111902

Castaneda

McLaren

Tavaziva

, et al. Biofilm antimicrobial susceptibility increases with antimicrobial exposure time. Clin Orthop Relat Res 2016; 474(7): 1659–1664. https://doi.org/10.1007/s11999-016-4700-z

10.

Ando

Miyahara

Hanada

, et al. Effective biofilm eradication in MRSA isolates with aminoglycoside-modifying enzyme genes using high-concentration and prolonged gentamicin treatment. Microbiol Spectr 2024; 12(10): e0064724. https://doi.org/10.1128/spectrum.00647-24

11.

Argenta

Morykwas

. Vacuum-assisted closure: a new method for wound control and treatment: clinical experience. Ann Plast Surg 1997; 38(6): 563–576.

12.

Kim

Attinger

Crist

, et al. Negative pressure wound therapy with instillation: review of evidence and recommendations. Wounds 2015; 27(12): S2–S19.

13.

Diehm

Loew

Will

, et al. Negative pressure wound therapy with instillation and dwell time (NPWTi-d) with V. A. C. VeraFlo in traumatic, surgical, and chronic wounds-A helpful tool for decontamination and to prepare successful reconstruction. Int Wound J 2020; 17(6): 1740–1749. https://doi.org/10.1111/iwj.13462, Epub 2020 Jul 27.

14.

Stewart

Clarke

Rovers

, et al. Preferred reporting items for systematic review and meta-analyses of individual participant data: the PRISMA-IPD statement. JAMA 2015; 313(16): 1657–1665. https://doi.org/10.1001/jama.2015.3656

15.

Singh

Goyal

Patel

. C-Reactive protein: clinical relevance and interpretation. In: StatPearls [internet]. Treasure Island (FL). StatPearls Publishing, 2025. Jan– [updated 2025 May 3; cited 2025 Aug 20]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK441843/

16.

Mount

Sinai.

C-reactive protein [internet]. Mount Sinai Health

System, 2025, [cited 2025 Aug 20]. Available from: https://www.mountsinai.org/health-library/tests/c-reactive-protein

17.

American College of

Clinical Pharmacy. Reference values for common laboratory tests. Pharmacotherapy Self-Assessment Program (PSAP), [Internet]. Lenexa (KS): ACCP; [cited 2025 Aug 20]. Available from: https://www.accp.com/docs/sap/Lab_Values_Table_PSAP.pdf

18.

Hieda

Choe

Maruo

, et al. Clinical outcomes of continuous local antibiotic perfusion in combination with debridement antibiotics and implant retention for periprosthetic hip joint infection. Sci Rep 2025; 15(1): 26017. https://doi.org/10.1038/s41598-025-11808-y

19.

Centers for Disease Control and Prevention (US) , National Healthcare Safety Network (NHSN). Chapter 9: Surgical Site Infection (SSI) Event. In: NHSN Patient Safety Component Manual [Internet]. CDC, 2025, p. 16. [cited 2025 Aug 19]. Available from: https://www.cdc.gov/nhsn/pdfs/pscmanual/9pscssicurrent.pdf

20.

Nagata

Yamada

Shinozaki

, et al. Effect of antimicrobial prophylaxis duration on health care–associated infections after clean orthopedic surgery: a cluster randomized trial. JAMA Netw Open 2022; 5(4): e226095. https://doi.org/10.1001/jamanetworkopen.2022.6095

21.

Davis

Metcalf

Clark

, et al. Predictors of treatment success after periprosthetic joint infection: 24-month follow-up from a multicenter prospective observational cohort study of 653 patients. Open Forum Infect Dis 2022; 9(3): ofac048. https://doi.org/10.1093/ofid/ofac048

22.

Pulido

Ghanem

Joshi

, et al. Periprosthetic joint infection: the incidence, timing, and predisposing factors. Clin Orthop Relat Res 2008; 466(7): 1710–1715. https://doi.org/10.1007/s11999-008-0209-4

23.

Metoxen

Ferreira

Zhang

, et al. Hospital readmissions after total joint arthroplasty: an updated analysis and implications for value-based care. J Arthroplast 2023; 38(3): 431–436. https://doi.org/10.1016/j.arth.2022.09.015

24.

Page

McKenzie

Bossuyt

, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021; 372: n71. https://doi.org/10.1136/bmj.n71

25.

Gagnier

Kienle

Altman

, et al. CARE group. The CARE guidelines: consensus-based clinical case reporting guideline development. Headache 2013; 53(10): 1541–1547. https://doi.org/10.1111/head.12246

26.

Munn

Barker

Moola

Tufanaru

Stern

McArthur

Stephenson

Aromataris

. Methodological quality of case series studies: an introduction to the JBI critical appraisal tool. JBI Evid Synth. Joanna Briggs Institute (JBI), 2020, 18(10), pp. 2127–2133. https://doi.org/10.11124/JBISRIR-D-19-00099

27.

Egger

Davey Smith

Schneider

, et al. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315(7109): 629–634. https://doi.org/10.1136/bmj.315.7109.629

28.

Parvizi

Zmistowski

Berbari

, et al. New definition for periprosthetic joint infection: from the workgroup of the musculoskeletal infection society. Clin Orthop Relat Res 2011; 469(11): 2992–2994. https://doi.org/10.1007/s11999-011-2102-9

29.

Parvizi

Tan

Goswami

, et al. The 2018 definition of periprosthetic hip and knee infection: an evidence-based and validated criteria. J Arthroplast 2018; 33(5): 1309–1314.e2. https://doi.org/10.1016/j.arth.2018.02.078

30.

Metsemakers

Morgenstern

McNally

, et al. Fracture-related infection: a consensus on definition from an international expert group. Injury 2018; 49(3): 505–510. https://doi.org/10.1016/j.injury.2017.08.040

31.

Stevens

Bisno

Chambers

, et al. Practice guidelines for the diagnosis and management of skin and soft tissue infections: 2014 update by the infectious diseases society of America. Clin Infect Dis 2014; 59(2): e10–e52. https://doi.org/10.1093/cid/ciu296

32.

Momodu II

Savaliya V.

Septic arthritis. In: StatPearls [Internet]. StatPearls Publishing, 2025. Jan– [updated 2023 Jul 3; cited 2025 Aug 20]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK538176/

33.

McNally

Sousa

Wouthuyzen-Bakker

, et al. The EBJIS definition of periprosthetic joint infection: A practical guide for clinicians. Bone Jt J. 2021; 103-B(1): 18–25. https://doi.org/10.1302/0301-620X.103B1.BJJ-2020-1381.R1

34.

Glaudemans

AWJM

Jutte

Cataldo

, et al. Consensus document for the diagnosis of peripheral bone infection in adults: a joint paper by the EANM, EBJIS, and ESR (with ESCMID endorsement). Eur J Nucl Med Mol Imaging 2019; 46(4): 957–970. https://doi.org/10.1007/s00259-019-4262-x

35.

Dindo

Demartines

Clavien

. Classification of surgical complications: a new proposal with evaluation in a cohort of 6336 patients and results of a survey. Ann Surg 2004; 240(2): 205–213. https://doi.org/10.1097/01.sla.0000133083.54934.ae

36.

Turnbull

. The empirical distribution function with arbitrarily grouped, censored and truncated data. J Roy Stat Soc B 1976; 38(3): 290–295, Available from. https://doi.org/10.1111/j.2517-6161.1976.tb01597.x

37.

Anderson-Bergman

icenReg: regression models for interval censored data in R. J Stat Softw 2017; 81(12): 1–23. https://doi.org/10.18637/jss.v081.i12

38.

Royston

Parmar

. The use of restricted mean survival time to estimate the treatment effect in randomized clinical trials when the proportional hazards assumption is in doubt. Stat Med 2011; 30(19): 2409–2421. https://doi.org/10.1002/sim.4274, Epub 2011 May 25.

39.

Uno

Claggett

Tian

, et al. Moving beyond the hazard ratio in quantifying the between-group difference in survival analysis. J Clin Oncol 2014; 32(22): 2380–2385. https://doi.org/10.1200/JCO.2014.55.2208, Epub 2014 Jun 30.

40.

Giordano

Giannoudis

. Biofilm formation, antibiotic resistance, and infection (BARI): the triangle of death. J Clin Med 2024; 13(19): 5779. https://doi.org/10.3390/jcm13195779

41.

Shimada

Ochiai

Hashimoto

, et al. Continuous local antibiotic perfusion technique for surgical site infections after shoulder surgery. JSES Rev Rep Tech 2024; 4(3): 419–423. https://doi.org/10.1016/j.xrrt.2024.04.013

42.

Yoon

Lee

Park

, et al.

Cutibacterium acnes in shoulder surgery: is it a significant risk factor for postoperative infection?

Clin Orthop Surg 2024; 16(6): 845–853. https://doi.org/10.4055/cios23371

43.

Vilchez

Escudero-Sanchez

Fernandez-Sampedro

, et al. Prosthetic shoulder joint infection by cutibacterium acnes: does rifampin improve prognosis? A retrospective, multicenter, observational study. Antibiotics (Basel) 2021; 10(5): 475. https://doi.org/10.3390/antibiotics10050475

44.

Schoenfeld

. Partial residuals for the proportional hazards regression model. Biometrika 1982; 69(1): 239–241. https://doi.org/10.1093/biomet/69.1.239

45.

Zenke

Motojima

Ando

, et al. DAIR in treating chronic PJI after total knee arthroplasty using continuous local antibiotic perfusion therapy: a case series study. BMC Musculoskelet Disord 2024; 25(1): 36. https://doi.org/10.1186/s12891-024-07165-y

46.

Yamaguchi

Ueda

Ichiseki

, et al. Effective management of methicillin-resistant shoulder septic arthritis using continuous local antibiotic perfusion: a case study and long-term follow-up. Am J Case Rep 2024; 25: e944491. https://doi.org/10.12659/AJCR.944491

47.

Fujihara

Uchibori

Yoshimoto

, et al. Prognostic factors influencing the occurrence of drug-induced renal dysfunction during continuous local antibiotic perfusion therapy. J Orthop Sci 2025; 31(1): 226–229. https://doi.org/10.1016/j.jos.2025.05.007

48.

Schrank

Jozefowski

Zura

, et al. PREP-IT Investigators. Local antibiotics and the risk of antimicrobial resistance in extremity fractures complicated by fracture-related infection. J Bone Joint Surg Am 2025; 107(Suppl 1): 28–35. https://doi.org/10.2106/JBJS.24.01178

49.

Charette

Melnic

. Two-stage revision arthroplasty for the treatment of prosthetic joint infection. Curr Rev Musculoskelet Med 2018; 11(3): 332–340. https://doi.org/10.1007/s12178-018-9495-y

50.

Singer

Merz

Frommelt

, et al. High rate of infection control with one-stage revision of septic knee prostheses excluding MRSA and MRSE. Clin Orthop Relat Res 2012; 470(5): 1461–1471. https://doi.org/10.1007/s11999-011-2174-6

51.

Lum

Holland

Meehan

. Systematic review of single stage revision for prosthetic joint infection. World J Orthop 2020; 11(12): 559–572. https://doi.org/10.5312/wjo.v11.i12.559

52.

Gupta

Shahban

Petrie

, et al.

DAIR for periprosthetic joint infections—one week to save the joint?

Arthroplasty 2024; 6(1): 61. https://doi.org/10.1186/s42836-024-00282-y

53.

Scholten

Hannink

Somford

, et al. The value of repeated debridement, antibiotics, and implant retention (DAIR) for early periprosthetic joint infection. J Bone Jt Infect 2025; 10(4): 207–215. https://doi.org/10.5194/jbji-10-207-2025

54.

Rombach

Zambellas

, et al. Oral versus intravenous antibiotics for bone and joint infection. N Engl J Med 2019; 380(5): 425–436. https://doi.org/10.1056/NEJMoa1710926

55.

Klemm

. Gentamycin-PMMA-Kugeln in der Behandlung abszedierender Knochen-und Weichteilinfektionen. Zentralbl Chir 1979; 104(14): 934–942, (in German).

56.

Bor

Dujovny

Rinat

, et al. Treatment of chronic osteomyelitis with antibiotic-impregnated polymethyl methacrylate (PMMA)—The Cierny approach: is the second stage necessary? BMC Musculoskelet Disord 2022; 23(1): 38. https://doi.org/10.1186/s12891-021-04979-y

57.

Piovan

Farinelli

Screpis

, et al. The role of antibiotic calcium sulfate beads in acute periprosthetic knee infection: a retrospective cohort study. Arthroplasty 2022; 4(1): 42. https://doi.org/10.1186/s42836-022-00139-2

58.

Tarity

Xiang

Jones

, et al. Do antibiotic-loaded calcium sulfate beads improve outcomes after debridement, antibiotics, and implant retention? A matched cohort study. Arthroplast Today 2022; 14: 90–95. https://doi.org/10.1016/j.artd.2022.01.023

59.

Sigmund

Palmer

AJR

Hotchen

, et al. The use of antibiotic-loaded calcium sulphate beads in DAIR for periprosthetic infections: a retrospective comparative cohort on outcome. Acta Orthop 2024; 95: 707–714. https://doi.org/10.2340/17453674.2024.42360

60.

Tissingh

Marais

Loro

, et al.

Management of fracture-related infection in low resource settings: how applicable are the current consensus guidelines?

EFORT Open Rev 2022; 7(6): 422–432. https://doi.org/10.1530/EOR-22-0031