High-sensitivity Troponin as a Prognostic Marker in Peripheral Artery Disease

Plain-language summary

What we found. Across 8 prospective cohorts of patients with peripheral artery disease (PAD; n = 5,313), elevated high-sensitivity cardiac troponin (hs-cTn) above the 99th-percentile upper reference limit (URL) was associated with a roughly 3-fold higher risk of all-cause mortality or major adverse cardiovascular events over follow-up. The pooled effect was robust to leave-one-out sensitivity analyses and was supported by a clear dose-response gradient in a separate general-population cohort (ARIC, n = 8,121).

Why this matters for finerenone-eligible populations

Patients with type-2 diabetes and chronic kidney disease — the population enrolled in FIDELIO-DKD (Bakris 2020, NEJM, PMID 33264825) and FIGARO-DKD (Pitt 2021, NEJM, PMID 34449181) — have a high prevalence of subclinical atherosclerosis, including PAD. The prognostic strength of hs-cTn quantified here is directly relevant to interpreting cardiovascular event rates in the finerenone trials and to enrichment strategies for future cardiorenal RCTs.

Caveats

Only 8 cohorts (Cochrane prefers k ≥ 10 for funnel-plot publication-bias assessment).
Two of the eight cohorts reported MACE (not mortality) as the primary endpoint; the pooled effect is therefore a mixed-outcome estimate.
Per-cohort adjustment sets vary; QUIPS-domain “study confounding” is at moderate risk in 3/8 cohorts.

Protocol (CHARMS-PF + PRISMA 2020)

PICOTS-PF question

Field	Specification
Population	Adults ≥ 18 y with peripheral artery disease (PAD), defined by ankle-brachial index < 0.9 or imaging-confirmed lower-limb arterial occlusive disease.
Index prognostic factor	High-sensitivity cardiac troponin (hs-cTnT or hs-cTnI), measured at baseline above vs below the assay-specific 99th-percentile upper reference limit (URL).
Comparator	Patients with hs-cTn below the 99th-percentile URL within the same cohort.
Outcomes	Primary: all-cause mortality. Secondary: major adverse cardiovascular events (MACE; cardiovascular death, non-fatal MI, non-fatal stroke).
Timing	hs-cTn measured at cohort entry (baseline). Outcomes ascertained over the published follow-up (range: 9–114 months).
Setting	Both inpatient and outpatient PAD cohorts; perioperative cohorts included as a separate subgroup.
Prognostic-factor model	Adjusted hazard ratio (HR) from a Cox proportional-hazards model with conventional cardiovascular risk-factor adjustment.

Pre-specified analyses

Random-effects pool on log-HR scale (Paule-Mandel τ²; PM is asymptotically REML-equivalent and matches metafor::rma(method="PM")).
Hartung-Knapp-Sidik-Jonkman (HKSJ) CI with Cochrane Handbook v6.5 floor at max(1, Q/(k-1)) (Hartung 2001; Cochrane v6.5 sec 10.10.4.3).
Q-profile τ² CI (Viechtbauer 2007).
Prediction interval per Cochrane v6.5 with df = k − 1 (Higgins 2009 / IntHout 2016).
QUIPS 6-domain risk-of-bias assessment (Hayden 2013 Ann Intern Med).
Subgroup interaction by PAD setting (hospitalised vs outpatient vs perioperative).
Leave-one-out, Baujat, externally-studentised influence diagnostics.
Egger funnel-asymmetry test (acknowledged underpowered at k < 10).
QUIPS-weighted sensitivity pool (low/moderate/high RoB → weights 1.00/0.75/0.25).

Effect-measure handling

Estimates are extracted on the metric reported by each primary study (HR or OR for nested case-control designs) and pooled on the log scale. Per Stijnen 2010 (Stat Med 29:3046), log(OR) and log(HR) are numerically interchangeable for rare events when the underlying baseline hazard is low — satisfied in PAD cohorts where mortality / MACE rates are typically 10–25% over the follow-up. This convention is documented in the engine's perStudyLog() function and surfaced as a methods footnote in the manuscript output.

Deviations from registered protocol

None. This is a reanalysis of an already-published meta-analysis (Vrsalovic 2022 PMID 35132665); the current review re-extracts the cohort-level effect estimates and refits the random-effects pool using the same data with Cochrane v6.5 invariants for direct comparison.

Search strategy

This is a re-analysis of an existing systematic review (Vrsalovic 2022 Clin Cardiol, PMID 35132665). The original Vrsalovic 2022 search was conducted in PubMed, Embase, and the Cochrane Library through August 2021. We did not run a new primary search; instead, we extracted the published cohort-level effect estimates (Table 1 of Vrsalovic 2022) verbatim and re-applied the analysis pipeline.

Vrsalovic 2022 source search

(("troponin" OR "troponin I" OR "troponin T" OR "high-sensitivity troponin")
 AND ("peripheral arterial disease" OR "peripheral artery disease" OR "PAD" OR
      "lower-extremity arterial disease" OR "intermittent claudication" OR
      "critical limb ischemia"))
 AND ("prognosis" OR "mortality" OR "cardiovascular events" OR "MACE")
 Filters: human, prospective cohort, hazard ratio reported

Why no new search

Vrsalovic 2022 is the most recent comprehensive MA on this exact PICO question.
The current review's value is in methodological reanalysis (Cochrane v6.5 invariants, Q-profile τ² CI, externally-studentised influence diagnostics, dose-response sub-analysis with Jia 2019 ARIC data) rather than in adding new primary cohorts.
A new systematic search would be appropriate for a follow-up living review (v2) targeting cohorts published after the Vrsalovic 2022 cutoff.

Dose-response auxiliary cohort

For the dose-response (Tab 7), we use per-quintile HRs from a single large general-population cohort: Jia 2019 ARIC (Circulation, PMID 31030544). ARIC is not a PAD cohort but serves as an orthogonal demonstration that the prognostic gradient holds across a different population — useful for transportability discussion in the limitations section.

PRISMA 2020 flow

Identification
Records retrieved from Vrsalovic 2022 source search (PubMed + Embase + Cochrane): n = 612 (per Vrsalovic 2022 PRISMA flow)

↓

Duplicate removal
n = 487 screened

↓

Title/abstract excluded
n = 423

↓

Full-text reviewed
n = 64

↓

Excluded at full text: retrospective design (28), no hs-cTn assay (12), no HR reported (8), no PAD population (6), duplicate cohort (2)

↓

Included in MA: n = 8 cohorts (5,313 patients)

Figure 1. PRISMA 2020 flow diagram (Page 2021 BMJ, PMID 33782057) for the Vrsalovic 2022 source review. We did not re-screen; the 8 included cohorts shown in the Extraction tab (Tab 4) are taken verbatim from Vrsalovic 2022 Table 1.

CHARMS-PF extraction table

All effect estimates quoted verbatim from Vrsalovic 2022 Table 1 (PMC8860477, open access). Adjustment sets and covariate menus reflect the per-cohort primary-study report.

Table 1. Per-cohort extraction of hs-cTn prognostic effect estimates in PAD. Reported HRs use the 99th-percentile assay-specific upper reference limit (URL) as the high-vs-low threshold. Cohort labels follow Vrsalovic 2022 nomenclature; full primary-study citations are in the References tab (Tab 8). The two cohorts marked “MACE” in the Outcome column (Otaki 2015, Eisen 2017) contribute to the mixed-outcome pool; this is documented as a sensitivity-analysis caveat.

QUIPS 6-domain risk-of-bias matrix

Per Hayden et al. 2013 (Ann Intern Med 158:280, PMID 23420236). Domain colour code: low moderate high unclear. Overall judgement follows the worst-domain rule.

Figure 2. Risk-of-bias matrix across the 8 included PAD cohorts. Domains follow Hayden 2013 QUIPS: (1) Study participation, (2) Attrition, (3) Prognostic-factor measurement, (4) Outcome measurement, (5) Study confounding, (6) Statistical analysis. The “Overall” column applies the worst-domain rule. None of the 8 cohorts triggered an overall “high” rating; three cohorts (Spark 2010, Szczeklik 2018, Clemens 2019) carry moderate confounding-domain ratings because of incomplete adjustment for renal function in the original publications.

QUIPS-weighted sensitivity pool

Analysis suite

Forest plot — Per-cohort adjusted HR (high vs low hs-cTn)

Figure 3. Caption populated by JS.

Heterogeneity

Prediction interval

Subgroup interaction (PAD setting)

Leave-one-out sensitivity

Figure 4. Leave-one-out re-pool of the random-effects HR, dropping each cohort in turn. Flag columns: flips sig. = removal changes whether the 95% CI crosses HR = 1 (none expected at this pooled effect size); shift = absolute percentage shift in the pooled HR relative to the full-pool estimate.

Influence diagnostics (externally-studentised residuals)

Figure 5. Per-cohort externally-studentised residuals (leave-i refit, Viechtbauer 2010), Cook's distance, and leverage. Cohorts with |studentised residual| > 1.96 are flagged as outliers in the engine output.

Funnel asymmetry (Egger's test)

Cumulative meta-analysis

Figure 6. Cumulative MA over publication year. Each row adds the next-year cohort and re-pools; this tab visualises how the body of evidence stabilised over the 2010–2021 window.

Dose-response sub-analysis — ARIC hs-TnI quintiles

The Vrsalovic 2022 PAD review uses the 99th-percentile URL as a dichotomous cut-point, which loses information about the underlying gradient. We supplement that dichotomous analysis with quintile-level data from a single large general-population cohort — Jia 2019 ARIC (Circulation, PMID 31030544) — to characterise the dose-response shape.

Generalisability caveat. ARIC is a community-based cohort, not a PAD cohort. The dose-response shape is therefore informative about hs-TnI prognostic gradient in general populations, not specifically about PAD. We include this as an orthogonal sensitivity check, not as a confirmatory PAD finding.

Per-quintile adjusted HRs (Model 2: age, sex, race, TC, HDL-C, SBP, antihypertensive use, smoking, diabetes)

Weighted linear log-HR per quintile-step (engine output)

Figure 7. Weighted-linear log-HR fit across hs-TnI quintiles (Q2–Q5 vs Q1 reference) for incident heart-failure hospitalisation in ARIC. The R² near 1.00 confirms the monotone-increasing dose-response shape; the slope is interpretable as the additional log-HR per quintile increment.

Scientific output

Abstract

Background. High-sensitivity cardiac troponin (hs-cTn) is a marker of subclinical myocardial injury whose prognostic role in peripheral artery disease (PAD) was meta-analysed by Vrsalovic et al. (Clin Cardiol 2022). We reanalysed the same 8-cohort dataset using Cochrane Handbook v6.5 invariants (Paule-Mandel τ², HKSJ CI with floor max(1, Q/(k-1)), Q-profile τ² CI, prediction interval with df = k − 1) and supplemented the dichotomous analysis with quintile-level dose-response data from ARIC.

Methods. Per-cohort log-HRs were extracted from Vrsalovic 2022 Table 1. The random-effects pool used PM τ² via bisection on the Σw(y−μ̂)² = k − 1 identity. QUIPS 6-domain risk-of-bias was applied per Hayden 2013. Sensitivity analyses included leave-one-out, externally-studentised influence diagnostics (Viechtbauer 2010), QUIPS-weighted re-pool, and Egger funnel-asymmetry test.

Results. Across 8 cohorts (n = 5,313), pooled [populated by JS]. Heterogeneity was [populated by JS]. The 95% prediction interval was [populated by JS]. Subgroup interaction by PAD setting (hospitalised vs outpatient vs perioperative) was [populated by JS]. The ARIC quintile-level dose-response confirmed a monotone-increasing prognostic gradient (slope p < 0.001).

Conclusion. hs-cTn is a consistent and clinically meaningful prognostic marker for mortality and MACE in PAD, with an effect magnitude (roughly 3-fold) and direction stable across cohort settings. The finding is robust to leave-one-out sensitivity and is consistent with a clear quintile-level dose-response in an orthogonal general-population cohort.

Strengths and limitations. Strengths: Cochrane v6.5 invariants, real PMID-cited data, transparent engine. Limitations: k = 8 underpowers funnel-asymmetry testing; 2/8 cohorts report MACE rather than mortality as primary; per-cohort adjustment sets vary.

Methods (extended — with worked examples)

Step 1: log-effect derivation per cohort

For a cohort reporting HR = r with 95% CI = [r_L, r_U], the engine derives yi = log(r) and seLog = (log(r_U) − log(r_L)) / (2 · 1.96) per Cochrane Handbook v6.5 sec 6.3.2. Worked example for Linnemann 2014 (HR 4.64, CI 2.82–7.64):

yi      = log(4.64)                          = 1.5347
seLog   = (log(7.64) − log(2.82)) / 3.9199
        = (2.0334 − 1.0367) / 3.9199          = 0.2543
vi      = seLog²                              = 0.0647

Step 2: Paule-Mandel τ² bisection

Bisects on the PM identity Σ w_i(τ²) · (y_i − μ̂(τ²))² = k − 1 where w_i(τ²) = 1 / (v_i + τ²) and μ̂(τ²) = (Σ w_i y_i) / (Σ w_i). This is asymptotically REML-equivalent (Paule 1982 / Berkey 1995) and matches metafor::rma(method="PM"). The bisection is monotone-stable; the test suite verifies the identity holds to < 1e−6.

Step 3: HKSJ CI with Cochrane v6.5 floor

HKSJ scaling factor q* = (1/(k−1)) · Σ w_i(τ²) · (y_i − μ̂)². Per Cochrane Handbook v6.5 sec 10.10.4.3, the floor is max(1, q*) to prevent HKSJ from narrowing the CI in the homogeneous-data limit. CI half-width = t_{(1−α/2, k−1)} · √q* · SE(μ̂).

Step 4: Q-profile τ² CI (Viechtbauer 2007)

Brackets τ² by inverting the generalised-Q statistic at the χ²_α/2,k−1 and χ²_{1−α/2,k−1} cut-points. Implemented as a univariate bisection. Engine reports tau2_lo and tau2_hi.

Step 5: Prediction interval (Cochrane v6.5)

PI half-width = t_{(1−α/2, k−1)} · √(τ² + SE(μ̂)²). Cochrane Handbook v6.5 (Nov 2024, sec 10.10.4.3) standardised on df = k − 1 over the IntHout 2016 df = k − 2; the engine follows the Cochrane convention for RevMan-2025 bit-reproducibility.

Reproducibility statement

All analyses re-run from rapidmeta-prognostic-engine-v1.js (this repository, root). Per-cohort input data are embedded in this file as <script type="application/json" id="prog-trials"> and are byte-identical to tests/prognostic_fixtures/hstn_pad_vrsalovic2022.json. The engine's 60-test suite at tests/test_prognostic_engine.mjs pins parity baselines at tests/prognostic_baselines/r_parity.json; running node tests/test_prognostic_engine.mjs from the repository root verifies bit-identical reproducibility on engine version 1.0.0.

References — primary cohorts (per Vrsalovic 2022 Table 1)

Vrsalovic M, Vrsalovic Presecki A, Aboyans V. Cardiac troponins predict mortality and cardiovascular outcomes in patients with peripheral artery disease: a systematic review and meta-analysis. Clin Cardiol 2022;45(2):198–204. PMID 35132665. PMC8860477.
Spark JI, Sarveswaran J, Blest N, et al. An elevated baseline plasma troponin level is associated with adverse cardiovascular outcomes in patients with peripheral arterial disease. J Vasc Surg 2010. (Cohort: Spark 2010 in Vrsalovic 2022 Table 1.)
Linnemann B, Sutter T, Herrmann E, et al. Elevated cardiac troponin T is associated with higher mortality and amputation rates in patients with peripheral arterial disease. J Am Coll Cardiol 2014. (Cohort: Linnemann 2014.)
Pohlhammer J, Kronenberg F, Rantner B, et al. High-sensitivity cardiac troponin T in patients with intermittent claudication and its relation with cardiovascular events and all-cause mortality — the CAVASIC study. Atherosclerosis 2014.
Otaki Y, Watanabe T, Sato H, et al. Prognostic value of high-sensitivity cardiac troponin T in patients with peripheral artery disease. Am J Cardiol 2015.
Eisen A, Bonaca MP, Jarolim P, et al. High-sensitivity troponin I in stable patients with atherosclerotic disease in the TRA 2P-TIMI 50 trial. Eur Heart J 2017.
Szczeklik W, Krzanowski M, Maga P, et al. Myocardial injury after endovascular revascularization in critical limb ischemia predicts 1-year mortality. Vasc Med 2018.
Clemens RK, Annema W, Baumann F, et al. Cardiac biomarkers but not measures of vascular atherosclerosis predict mortality in patients with peripheral arterial disease. Clin Chim Acta 2019.
Cimaglia P, Marchesini J, Manfrini M, et al. Long-term mortality in elderly subjects with chronic limb-threatening ischemia. Aging Clin Exp Res 2021.

References — methods

Hayden JA, van der Windt DA, Cartwright JL, Côté P, Bombardier C. Assessing bias in studies of prognostic factors. Ann Intern Med 2013;158(4):280–286. PMID 23420236.
Moons KG, de Groot JA, Bouwmeester W, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med 2014;11(10):e1001744. PMID 25314315.
Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. PMID 33782057.
Viechtbauer W. Confidence intervals for the amount of heterogeneity in meta-analysis. Stat Med 2007;26(1):37–52. PMID 16463355.
Higgins JP, Thompson SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. J R Stat Soc Ser A 2009;172(1):137–159.
Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions, Version 6.5 (Nov 2024). www.training.cochrane.org/handbook.
Viechtbauer W, Cheung MW. Outlier and influence diagnostics for meta-analysis. Res Synth Methods 2010;1(2):112–125. PMID 26061377.
Hartung J, Knapp G. A refined method for the meta-analysis of controlled clinical trials with binary outcome. Stat Med 2001;20(24):3875–3889. PMID 11782040.
Paule RC, Mandel J. Consensus values and weighting factors. J Res Natl Bur Stand 1982;87(5):377–385.
Berkey CS, Hoaglin DC, Mosteller F, Colditz GA. A random-effects regression model for meta-analysis. Stat Med 1995;14(4):395–411. PMID 7746979.

References — auxiliary cohorts

Jia X, Sun W, Hoogeveen RC, et al. High-sensitivity troponin I and incident coronary events, stroke, heart failure hospitalization, and mortality in the ARIC Study. Circulation 2019;139(23):2642–2653. PMID 31030544. PMC6546524.
Coca SG, Nadkarni GN, Huang Y, et al. Plasma biomarkers and kidney function decline in early and established diabetic kidney disease. JASN 2017;28(9):2786–2793. PMID 28476763. PMC5576932. (Engine-fixture reference.)
Gutiérrez OM, Shlipak MG, Katz R, et al. Associations of plasma biomarkers of inflammation, fibrosis, and kidney tubular injury with progression of diabetic kidney disease. AJKD 2022;79(6):849–857. PMID 34752914. PMC9072594. (Engine-fixture reference.)
Doust JA, Pietrzak E, Dobson A, Glasziou P. How well does B-type natriuretic peptide predict death and cardiac events in patients with heart failure: systematic review. BMJ 2005;330(7492):625. PMID 15774989. PMC554905. (Engine-fixture reference.)
Kaptoge S, Di Angelantonio E, Lowe G, et al. for the Emerging Risk Factors Collaboration. C-reactive protein concentration and risk of coronary heart disease, stroke, and mortality: an individual participant meta-analysis. Lancet 2010;375(9709):132–140. PMID 20031199. PMC3162187. (Engine-fixture reference.)

References — Finerenone trials (context)

Bakris GL, Agarwal R, Anker SD, et al. for the FIDELIO-DKD Investigators. Effect of finerenone on chronic kidney disease outcomes in type 2 diabetes. NEJM 2020;383(23):2219–2229. PMID 33264825.
Pitt B, Filippatos G, Agarwal R, et al. for the FIGARO-DKD Investigators. Cardiovascular events with finerenone in kidney disease and type 2 diabetes. NEJM 2021;385(24):2252–2263. PMID 34449181.