Background: Evidence for neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and lymphocyte-to-monocyte ratio (LMR) in primary bone sarcoma is dominated by osteosarcoma and has been affected by inconsistent endpoint and effect-size extraction. We re-extracted eligible estimates from the primary reports and evaluated their independent prognostic associations. Methods: MEDLINE, Embase, Web of Science, Scopus, and Cochrane CENTRAL were searched for cohort studies reporting pretreatment NLR, PLR, or LMR in primary bone sarcoma. The primary analysis required clinically multivariable-adjusted, categorical high-versus-low hazard ratios (HRs). Overall survival (OS) was analyzed separately from progression-free survival (PFS), disease-free survival (DFS), event-free survival (EFS), and disease-specific mortality. Random-effects models used restricted maximum likelihood (REML) estimation and Hartung-Knapp confidence intervals. Histology- and Newcastle-Ottawa Scale (NOS)-based subgroup analyses were prespecified when at least two studies were available. Results: Ten studies remained in the systematic review, and nine contributed at least one source-verifiable endpoint-specific adjusted estimate to quantitative synthesis. Seven NLR-OS cohorts yielded a pooled HR of 1.64 (95% CI 1.15–2.33; I² = 81.4%). The osteosarcoma-only subgroup remained significant (five cohorts; HR 1.70, 95% CI 1.03–2.81; I² = 86.6%). Excluding Huang X 2019 from the strict dataset yielded HR 1.84 (95% CI 1.44–2.36; I² = 0%); the reviewer-specified comparison against the previous eight-estimate set, retaining Yang Q 2020 only as a source-reported sensitivity input, yielded HR 1.90 (95% CI 1.47–2.45; I² = 5.9%). PLR was not associated with OS (four cohorts; HR 1.07, 95% CI 0.67–1.72; I² = 56.0%). LMR showed a non-significant protective trend (three cohorts; HR 0.70, 95% CI 0.38–1.27; I² = 0%). Conclusions: Higher pretreatment NLR is associated with poorer OS, including in the strict osteosarcoma-only analysis. The direction remained stable in every leave-one-out analysis, while the marked reduction in heterogeneity after excluding Huang X 2019 identifies one dominant statistical influence. Retrospective designs and residual heterogeneity limit clinical application. PLR and LMR findings remain inconclusive, and whether NLR-guided management improves outcomes is untested.