Lymphocyte-to-monocyte ratio predicts survival in localized osteosarcoma in Vietnamese patients

Summary

Inflammatory markers such as the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and lymphocyte-to-monocyte ratio (LMR), alongside hemoglobin (Hgb) levels, have been proposed as prognostic tools in osteosarcoma, though their reliability remains unclear. We retrospectively evaluated the prognostic significance of these pre-operative markers in a cohort of 100 Vietnamese patients with stage IB-IIB osteosarcoma, specifically focusing on the post-neoadjuvant, pre-operative window, a clinically critical yet underexplored time frame.

In the primary dichotomized multivariable analysis, LMR emerged as an independent predictor of poor overall survival, whereas PLR showed only univariable associations, and NLR and Hgb were not statistically significant. While LMR demonstrated robust prognostic value for overall survival, it failed to predict disease progression within 24 months. To address cutpoint bias, sensitivity analyses modeling these markers as continuous variables combined with internal bootstrap validation (1000 resamples) confirmed parameter stability only for LMR, but not PLR and NLR. Reflecting these continuous trends, the final multivariable model achieved an apparent C-index of 0.729, which adjusted to an optimism-corrected C-index of 0.628. External literature-derived median cutoffs did not show significant associations when applied to this post-neoadjuvant cohort, highlighting key population- and timing-specific limitations. Although these peripheral blood markers demonstrate prognostic potential, our findings remain exploratory, and validation in independent multi-center cohorts with longitudinal prospective evaluation is required before their clinical utility can be established.

Graphical abstract

Highlights

• •Post-neoadjuvant markers may reflect the host’s immune recovery and resilience
• •Prognostic performance varied substantially across NLR, PLR, LMR, and Hgb
• •LMR predicted overall survival but not 24-month disease progression
• •Prognostic value may be population- and timing-specific

Health sciences

Introduction

Osteosarcoma, though rare among all cancers, is the most common primary bone malignancy in adolescents. The global incidence of this cancer is approximately 3.3–4.0 cases per million people per year, with slightly lower rates in Asian populations.¹ The highly aggressive nature of osteosarcoma complicates the diagnosis and management of patients. Survival outcomes for the disease are also strongly dependent on the stage of diagnosis, with the highest 5-year survival rate (70–77%) observed in localized cases^2,3 but dropping significantly in patients with metastatic disease (24–36%).^4,5

The care and follow-up for patients with osteosarcoma are highly financially burdensome and require multiple time checks. An Australian study estimated the lifetime cost of surveillance after treatment for high-grade extremity sarcoma to be approximately US$4,032 for CT-only protocols and US$10,654 when MRI was added, corresponding to annual costs of roughly US$400–1,000 per patient.⁶ This estimation is similar in osteosarcoma. The most important factor that influences patients’ adherence to follow-up protocol is the cost.⁷ Therefore, to enhance treatment outcomes, more accessible, affordable, and easy-to-obtain tests, such as complete blood count and basic metabolic panel, are important in bringing patients back to clinics.

Over the last decades, a growing body of studies has explored the prognostic significance of systemic inflammatory biomarkers such as neutrophil-to-lymphocyte ratio (NLR), lymphocyte-to-monocyte ratio (LMR), and platelet-to-lymphocyte ratio (PLR) in various malignancies, including soft tissue carcinomas, breast, gastric, colorectal, and lung cancers.^8,9,10,11,12 In osteosarcoma, these markers have also gained attention due to their accessibility and potential to reflect tumor-related immune responses. For instance, elevated pre-treatment NLR was associated with poorer overall survival (OS) in French patients.¹³ Similarly, high PLR and low LMR were linked to worse outcomes in Chinese patients.¹⁴ However, most osteosarcoma biomarker studies have been concentrated in China, the USA, and Japan,¹⁵ with limited data from Southeast Asia, underscoring the need for studies involving more diverse ethnic and geographic populations to validate or even challenge current findings.

While biomarkers may provide prognostic information, their predictive value can vary across different stages and treatment courses of the disease. Most previous studies have primarily focused on biomarker measurements obtained before neoadjuvant chemotherapy.^16,17 However, several tumor suppressor genes such as P16, P53, and PTEN also have prognostic significance but only after chemotherapy.¹⁸ These findings suggest that systemic therapy may alter the biological relevance of prognostic markers over time. Since neoadjuvant chemotherapy can modify the systemic immune and hematologic environment, post-neoadjuvant inflammatory markers may reflect not only residual tumor biology, but also the host response to treatment-related stress and immune recovery. This distinction represents a clinically relevant yet underexplored research gap.

Our study aims to evaluate the prognostic significance of post-neoadjuvant pre-operative hematological biomarkers in a Vietnamese osteosarcoma cohort.

Results

Patient selection

Among 321 patients with malignant bone tumors who underwent surgery at our hospital from 2019, we have a final enrolled cohort of 100 patients who meet the inclusion criteria and did not violate any exclusion criteria. The patient selection flowchart is described in Figure 1.

Figure 1.

Patient selection flowchart

∗Records before 2022 were excluded due to incomplete laboratory data available on digital medical records. ∗∗Patients were included only if treated ≥1 year before the study to ensure ≥12 months of follow-up for surviving patients.

Patient clinico-pathological characteristics

A total of 100 Vietnamese patients diagnosed with osteosarcoma were included in this retrospective study. The cohort comprised 62 females (62%) and 38 males (38%), resulting in a female-to-male ratio of approximately 1.6:1. The majority of patients (72%) were adolescents under 18 years of age, with a mean age at diagnosis of 15.85 ± 7.16 years, Table 1.

Table 1.

Clinicopathological characteristics of 100 patients with localized osteosarcoma

All patient characteristics	n (%)	Mean (SD)	Median (range)
Total
Male	38 (38%)	–	–
Female	62 (62%)	–	–
Age at diagnosis		15.85 (7.16)	14.00 (IQR 7.00)
<18 years	72 (72%)	–	–
≥18 years	28 (28%)	–	–
Ennecking stage at diagnosis
IB	3 (3%)	–	–
IIA	7 (7%)	–	–
IIB	90 (90%)	–	–
Definitive histology
Conventional OS	94 (94%)	–	–
High-grade surface OS	2 (2%)	–	–
Parosteal OS	2 (2%)	–	–
Telangiectatic OS	2 (2%)	–	–
Chemotherapy
Neoadjuvant	99 (99%)	–	–
Adjuvant	90 (90%)	–	–
Both	89 (89%)	–	–
Overall survival (months)		23.02 (8.1)	–
Onset to diagnosis (months)	–	2.68 (3.26)	–
Diagnosis to the first surgery (months)	–	5.08 (3.26)	–
Progressed within 24 months	28 (28%)	–	–
Progression-free ≥ 24 months	43 (43%)	–	–
Progression-free, follow-up time ≤24 months	29 (29%)	–	–
Deaths	10 (10%)	–	–
Biomarkers
RBC (1012/L)	–	3.75 (0.67)	–
WBC (109/L)	–	7.08 (2.49)	–
NEU (109/L)	–	4.76 (2.16)	–
LYM (109/L)	–	1.48 (0.84)	–
MONO (109/L)	–	0.65 (0.37)	–
PLT (109/L)	–	236.35 (100.27)	–
Hgb (g/L)	–	109.32 (18.56)	110.00 (72.00–149.00)
Anemic	72 (72%)	–	–
Non-Anemic	28 (28%)	–	–
NLR	–	3.99 (2.53)	3.43 (0.65–11.07)
LMR	–	3.33 (3.32)	2.27 (0.63–26.25)
PLR	–	190.26 (115.07)	158.22 (32.11–701.92)

IQR: interquartile range.

At presentation, most patients (90%) were classified as Enneking stage IIB. Only 3 patients (3%) were stage IA, and 7 patients (7%) were stage IIA, Table 1. Conventional osteosarcoma was the predominant histopathological subtype, identified in 94 patients (94%). The remaining cases included high-grade surface osteosarcoma (2%), parosteal osteosarcoma (2%), and telangiectatic osteosarcoma (2%).

Neoadjuvant chemotherapy was administered to 99 patients (99%) prior to surgery, and 90 patients (90%) received adjuvant chemotherapy. Overall, 89 patients (89%) received both neoadjuvant and adjuvant chemotherapy.

The mean interval from symptom onset to diagnosis was 2.68 ± 3.26 months, while the mean time from diagnosis to first surgery was 5.08 ± 3.26 months. The mean OS for the cohort was 23.02 ± 8.10 months.

Within 24 months of diagnosis, disease progression occurred in 28 patients (28%). Meanwhile, 43 patients (43%) remained progression-free for at least 24 months. The remaining 29 patients (29%) had follow-up durations of less than 24 months and showed no evidence of progression at the time of analysis. Among those with early progression, 10 patients (10%) died due to disease-related complications.

Preoperative blood test results showed a median Hgb of 110.00 g/L (range, 72.00–149.00), NLR of 3.43 (range, 0.65–11.07), LMR of 2.27 (range, 0.63–26.25), and PLR of 158.22 (range, 32.11–701.92).

The Shapiro-Wilk test demonstrated that all biomarker data were non-normally distributed (p < 0.05) in both the 24-month progression and non-progression groups. Consequently, non-parametric statistical methods were employed for group comparisons. The Mann-Whitney U test revealed no statistically significant differences in preoperative biomarker levels between the two groups, with all p values exceeding 0.05 (Table 2). Specifically, the mean NLR was 3.93 ± 2.70 in the progression group and 3.98 ± 2.54 in the progression-free group (p = 0.865). The mean LMR was 2.75 ± 1.64 for patients with progression and 4.21 ± 4.57 for those without progression (p = 0.278). Similarly, the mean PLR was 206.99 ± 146.13 in both groups (p = 0.939).

Table 2.

Mann-Whitney U test for the 24-month progression and non-progression group

Biomarker (mean ± SD)	Progressed <24 months (n = 28)	Progression-free ≥ 24 months (n = 43)	p value	Test used
NLR	3.93 ± 2.7	3.98 ± 2.54	0.865	Mann-Whitney U
LMR	2.75 ± 1.64	4.21 ± 4.57	0.278	Mann-Whitney U
PLR	206.99 ± 146.13	188.13 ± 95.32	0.939	Mann-Whitney U
Hgb	108.96 ± 20.15	109.98 ± 17.29	0.888	Mann-Whitney U

Dichotomized survival analysis

OS and PFS of all patients

Across the study cohort, the Kaplan-Meier survival analysis showed a high OS rate throughout the follow-up period. The estimated OS at 12 months was 93.0% (95% CI: 88.1%–98.14%), declining to 89.4% (95% CI: 83.3%–95.88%) by 24 months and remaining stable through 36 months (Figure 2A).

Figure 2.

Kaplan-Meier survival analysis

(A) Overall survival.

(B) Progression-free survival.

For PFS, 93.0% (95% CI: 88.1%–98.1%) of patients were free from osteosarcoma-related progression at 12 months (Figure 2B). This probability decreased to 71.7% (95% CI: 62.5%–82.4%) at 24 months and further declined to 55.3% (95% CI: 42.8%–71.4%) at 36 months.

Hgb and anemia status

Preoperative anemia was present in 72 patients (72%) within the study cohort. ROC analysis of Hgb for predicting clinical events yielded an AUC of 0.52 (95% CI: 0.33–0.72, p = 0.817), with an optimal cutoff value of 114.5 g/L determined by the Youden index (Figure 3A, Table 3).

Figure 3.

Survival analysis of Hgb

(A) ROC curve analysis. Cut-off value: 114.5; AUC: 0.52 (0.33–0.72), p = 0.817.

(B) OS using Hgb cutoff 114.5 g/L.

(C) OS between anemic vs. non-anemic groups.

Table 3.

Overall survival ROC curve analysis of all biomarkers

Biomarker	Cut-off value	AUC (95% confidence interval, p value)	Sensitivity (%)	Specificity (%)
Hgb	114.5 g/L	0.52 (0.33–0.72, p = 0.817)	70.0	58.9
NLR	6.57	0.52 (0.29–0.74, p = 0.892)	40.0	84.4
LMR	1.615	0.72 (0.55–0.89, p = 0.009)	60.0	78.9
PLR	234.275	0.62 (0.39–0.84, p = 0.307)	50.0	78.9

AUC: area under the curve, Hgb: hemoglobin, NLR: neutrophil-to-lymphocyte ratio, LMR: lymphocyte-to-monocyte ratio, PLR: platelet-to-lymphocyte ratio.

Survival analyses conducted using both the Hgb cutoff and anemia status revealed no significant differences in outcomes between patients with high versus low Hgb levels, nor between anemic and non-anemic groups (log rank p = 0.1 and p = 0.42, respectively; Figures 3B and 3C). Hgb was not included in further analyses related to early disease progression.

NLR and survival outcomes

In the OS analysis, the optimal preoperative NLR cutoff was determined to be 6.57, corresponding to a sensitivity of 40% and a specificity of 84.4%. The AUC was 0.52 (95% CI: 0.29–0.74, p = 0.892) (Figure 4A; Table 3). Although patients with NLR values below 6.57 exhibited improved survival, this difference did not reach statistical significance (log rank p = 0.054) (Figure 4B).

Figure 4.

Survival analysis of NLR

(A) OS ROC curve analysis. Cut-off value: 6.57; AUC: 0.52 (0.29–0.74), p = 0.892.

(B) OS Survival curve.

(C) 24-month PFS ROC curve analysis. Cut-off value: 3.6; AUC: 0.49 (0.34–0.63), p = 0.865.

(D) 24-month PFS Survival curve.

In the 24-month PFS analysis, NLR showed limited discriminative ability, with an optimal cutoff of 3.6 (AUC = 0.49 (95% CI: 0.34–0.63, p = 0.865) (Figure 4C; Table 4), and no significant difference in survival between groups (log rank p = 0.22) (Figure 4D).

Table 4.

24-month progression survival ROC curve analysis of NLR, LMR, and PLR

Biomarker	Cut-off value	AUC (95% confidence interval, p value)	Sensitivity (%)	Specificity (%)
NLR	3.60	0.49 (0.34–0.63, p = 0.865)	50.0	62.8
LMR	3.77	0.58 (0.44–0.71, p = 0.264)	75.0	41.9
PLR	166.27	0.51 (0.36–0.65, p = 0.937)	53.6	58.1

LMR and survival outcomes

The optimal LMR cutoff was determined to be 1.615, with a sensitivity of 60% and a specificity of 78.9%. The AUC was 0.72 (95% CI: 0.55–0.89, p = 0.0093) (Figure 5A; Table 3). Patients with an LMR ≥1.615 had significantly better survival compared to those with LMR <1.615 (log rank p = 0.0039) (Figure 5B).

Figure 5.

Survival analysis of LMR

(A) OS ROC curve analysis. Cut-off value: 1.615; AUC: 0.72 (0.55–0.89), p = 0.0093.

(B) OS Survival curve.

(C) 24-month PFS ROC curve analysis. Cut-off value: 3.77; AUC: 0.58 (0.44–0.71), p = 0.264.

(D) 24-month PFS Survival curve.

In the 24-month PFS analysis, the optimal LMR cutoff was 3.77 (AUC = 0.58, 95% CI: 0.44–0.71, p = 0.264) (Figure 5C; Table 4); and this threshold did not significantly differentiate between groups (log rank p = 0.12) (Figure 5D).

PLR and survival outcomes

The optimal cutoff value for PLR in OS analysis was 234.275 (sensitivity: 50%, specificity: 78.9%), with an AUC of 0.62 (95% CI: 0.39–0.84, p = 0.307) (Figure 6A; Table 3). Patients with PLR less than 234.275 had significantly better survival compared to those with PLR ≥234.275 (log rank p = 0.033) (Figure 6B).

Figure 6.

Survival analysis of PLR

(A) OS ROC curve analysis. Cut-off value: 234.275; AUC: 0.62 (0.39–0.84), p = 0.307.

(B) OS Survival curve.

(C) 24-month PFS ROC curve analysis. Cut-off value: 166.27; AUC: 0.51 (0.36–0.65), p = 0.937.

(D) 24-month PFS Survival curve.

In the 24-month PFS analysis, PLR exhibited limited prognostic value. The optimal cutoff for this analysis was 166.27 (AUC = 0.51 (95% CI: 0.36–0.65, p = 0.937) (Figure 6C; Table 4), and no significant difference in PFS was observed between groups (log rank p = 0.13) (Figure 6D).

For biomarkers with AUC greater than 0.5, multivariable Cox proportional hazards models were performed to identify independent prognostic factors for event-free survival and PFS, adjusting for age at diagnosis, sex, chemotherapy status, and biomarker levels.

In univariable and multivariate analysis for OS, LMR <1.615 was proved to be an independent predictor of diminished overall survival with hazard ratios (HR) of 5.29 (95% CI: 1.49–18.81, p = 0.010) and 4.19 (95% CI: 1.10–15.92, p = 0.035), respectively (Table 5). While PLR≥ 234.275 was initially associated with poorer prognosis in univariate analysis (HR 3.53 (1.02–12.20, p = 0.046)), its association was attenuated and lost statistical significance in multivariable analysis (HR 2.23 (0.56–8.82, p = 0.253)) (Table 5). NLR was not a significant prognostic factor in either univariable (HR 2.35 (95% CI: 0.58–9.42), p = 0.229) or multivariable analysis (HR 2.42 (95% CI: 0.60–9.81), p = 0.214) (Table 5).

Table 5.

OS multivariable Cox proportional hazards model analysis

Variable		All	HR (univariable)	HR (multivariable)
Gender	female	38 (38.0)	–	–
	male	62 (62.0)	0.65 (0.19–2.26, p = 0.503)	0.91 (0.25–3.32, p = 0.883)
Age at diagnosis	<18	72 (72.0)	–	–
	≥18	28 (28.0)	1.82 (0.51–6.46, p = 0.353)	1.90 (0.51–7.08, p = 0.342)
NLR groups	<6.57	82 (82.0)	–	–
	≥6.57	18 (18.0)	3.23 (0.91–11.47, p = 0.069)	2.39 (0.60–9.49, p = 0.216)
LMR groups	≥1.615	75 (75.0)	–	–
	<1.615	25 (25.0)	5.29 (1.49–18.81, p = 0.010)	4.19 (1.10–15.92, p = 0.035)
PLR groups	<234.275	76 (76.0)	–	–
	≥234.275	24 (24.0)	3.53 (1.02–12.20, p = 0.046)	2.23 (0.56–8.82, p = 0.253)
Chemotherapy (neoadjuvant, adjuvant therapy)	both	89 (89.0)	–	–
	incomplete/missing	11 (11.0)	0.85 (0.11–6.71, p = 0.876)	0.96 (0.11–8.38, p = 0.972)

In the 24-month PFS analysis, none of the biomarkers showed significant associations (all p > 0.05), as shown in Table 6.

Table 6.

24-month PFS multivariable Cox proportional hazards model analysis

Variable	–	All	HR (univariable)	HR (multivariable)
Gender	female	30 (42.3)	–	–
–	male	41 (57.7)	1.03 (0.48–2.22, p = 0.943)	1.09 (0.50–2.39, p = 0.831)
Age at diagnosis	≥18 years	18 (25.4)	–	–
–	<18 years	53 (74.6)	0.67 (0.29–1.53, p = 0.342)	0.78 (0.33–1.88, p = 0.585)
Chemotherapy	incomplete/missing	8 (11.3)	–	–
–	both	63 (88.7)	1.06 (0.32–3.53, p = 0.925)	1.41 (0.41–4.94, p = 0.587)
NLR Groups	≥3.6	30 (42.3)	–	–
–	<3.6	41 (57.7)	0.62 (0.29–1.33, p = 0.221)	0.71 (0.32–1.6, p = 0.410)
LMR Groups	≥3.77	26 (36.6)	–	–
–	<3.77	45 (63.4)	1.94 (0.82–4.60, p = 0.131)	1.91 (0.78–4.70, p = 0.158)
PLR Groups	<166.27	38 (53.5)	–	–
–	≥166.27	33 (46.5)	1.80 (0.84–3.87, p = 0.132)	1.69 (0.73–3.9, p = 0.219)

Sensitivity analysis and internal validation

Sensitivity analysis

The prognostic value of the inflammatory biomarkers was reevaluated using continuous variable modeling (Table 7). For all three indices, the proportional hazards assumption was satisfied based on Schoenfeld residual testing (all Schoenfeld p > 0.05) (Table 7).

Table 7.

Sensitivity analysis and internal validation

Variable	Mean (SD)	Univariable HR (95% CI, p value)	Multivariable HR (95% CI, p value)	Bootstrap HR (95% CI, p value)	C-index	Schoenfeld (p)
Gender	–	–	0.92 (0.25–3.42, p = 0.904)	0.92 (0.17–7.99, p = 0.904)	–	–
Age at diagnosis	15.85 (7.16)	–	1.04 (0.96–1.12, p = 0.315)	1.04 (0.87–1.18, p = 0.315)	–	–
NLR	4.0 (2.5)	1.07 (0.84–1.35, p = 0.597)	0.91 (0.63–1.3, p = 0.608)	0.91 (0.51–1.24, p = 0.608)	0.519	0.259
LMR	3.3 (3.3)	0.49 (0.24–1.00, p = 0.049)	0.54 (0.26–1.09, p = 0.085)	0.54 (0.19–0.78, p = 0.085)	0.729	0.125
PLR	190.3 (115.1)	1.00 (1.00–1.01, p = 0.019)	1.00 (1.00–1.01, p = 0.068)	1.00 (1.00–1.01, p = 0.068)	0.606	0.619
Apparent AUC = 0.729; optimism-corrected AUC = 0.628

In the univariable model, both LMR and PLR demonstrated statistically significant associations with overall survival (OS). Specifically, for every one-unit increase in LMR, the risk of death decreased by 51% (HR = 0.49, 95% CI: 0.24–1.00, p = 0.049). Conversely, higher PLR was associated with a marginal but statistically significant increase in the hazard of death (HR = 1.00, 95% CI: 1.00–1.01, p = 0.019). Continuous NLR did not show a significant prognostic association in this setting (HR = 1.07, p = 0.597) (Table 7).

After adjusting for gender and continuous age at diagnosis in the multivariable analysis, the prognostic strength of these indices was attenuated. The association for LMR remained suggestive of a protective effect but did not reach the threshold for statistical significance (HR = 0.54, 95% CI: 0.26–1.09, p = 0.085), and PLR similarly exhibited a non-significant trend (HR = 1.00, p = 0.068). NLR remained non-significant (HR = 0.91, p = 0.608) (Table 7).

Internal validation

To assess parameter stability and account for potential overfitting, bootstrap internal validation was performed using 1000 resamples. The discriminative ability of each continuous biomarker model was measured using Harrell’s C-index. Among the three biomarkers, LMR showed the highest discriminative ability with an apparent C-index of 0.729, followed by PLR (0.606) and NLR (0.519). After bootstrap correction, the optimism-corrected C-index for the LMR model was 0.628, with an optimism of 0.101 (Table 7).

Bootstrap-derived 95% confidence intervals for all multivariable hazard ratios were directionally concordant with their asymptotic counterparts across all predictors. The bootstrap CI for LMR was 0.19–0.78 (HR = 0.54) and for PLR was 1.00–1.010 (HR = 1.00). The bootstrap CI for Gender was notably wider (95% CI_boot: 0.17–7.99, HR = 0.92). LMR and PLR’s bootstrap CIs remained consistent with the direction of effect seen in the primary analysis.

Discussion

Although Hgb, NLR, LMR, and PLR are routinely accessible and have been proposed as prognostic indicators in various malignancies, their role in osteosarcoma remains under active investigation.¹⁹ A recent comprehensive meta-analysis by Song et al. demonstrated that while elevated pre-treatment NLR and CRP are robust indicators of poor survival, pre-treatment LMR did not show a statistically significant pooled correlation with patient outcomes.²⁰ In contrast to previous studies that utilized pre-treatment markers to assess baseline tumor aggressiveness,^21,22,23 our study explores the correlation between inflammatory biomarkers and anemia collected in the period of post-neoadjuvant therapy and before surgery, an underappreciated time window, to predict the prognosis of patients with osteosarcoma. This time frame is biologically intuitive, as systemic inflammation at this juncture may reflect the host’s immune recovery and physiological resilience in addition to treatment response and residual tumor biology. The biological relevance of post-neoadjuvant inflammatory biomarkers may also be interpreted in the context of chemotherapy-associated molecular alterations in osteosarcoma. It has been shown that the prognostic significance of tumor suppressor pathways, including P16 and PTEN, becomes more apparent in post-chemotherapy resection specimens rather than in pretreatment biopsies, suggesting that neoadjuvant therapy reshapes tumor biology and treatment resistance profiles.¹⁸ Integrating genomic landscape, tumor-microenvironment and immune-related gene expressions exhibits prognostic and predictive value of immunotherapy and chemotherapy in soft-tissue sarcoma.^24,25,26

Demographically, our cohort exhibited a 62% female proportion and a minor inclusion of rare subtypes, including parosteal, telangiectatic, and high-grade surface osteosarcoma (6% total). This distribution likely reflects the specific referral patterns of our institution as one of the major osteosarcoma centers in Vietnam. Despite these demographic variations, the prognostic significance of LMR remained robust. Gender was controlled for as a covariate in our multivariable analysis, and the 94% predominance of conventional osteosarcoma ensured a high degree of pathological homogeneity. Thus, while these characteristics represent a pragmatic perspective of the Vietnamese patient population, they do not diminish the statistical integrity or the potential clinical utility of our findings.

An interesting observation in our study is that LMR demonstrated significant prognostic value for overall survival but failed to predict progression within 24 months in the dichotomized analysis. While patients with low LMR showed markedly poorer overall survival, no statistically significant association was observed between LMR and early disease progression. This contrast suggests that post-neoadjuvant LMR may reflect systemic host factors, such as immune recovery, physiological reserve, and hematologic recovery following treatment, influencing long-term survival rather than tumor aggressiveness alone.^27,28,29 Neoadjuvant chemotherapy can substantially alter circulating immune cell populations, and reduced lymphocyte levels after treatment may reflect impaired immune recovery, whereas increased monocyte levels may indicate persistent systemic inflammation and myeloid activation. Therefore, a low LMR may represent a host state associated with impaired immune restoration and increased vulnerability to treatment-related stress. These conditions could influence long-term survival by affecting treatment tolerance, recovery following therapy, and the ability to maintain control of residual disease. In contrast, early progression within 24 months is likely to be more strongly determined by intrinsic tumor biology, metastatic potential, and treatment response.^30,31 Biomarkers derived from peripheral blood counts may therefore have limited ability to capture the complex molecular determinants of early relapse. This distinction may explain why LMR showed prognostic value for overall survival but not for short-term progression outcomes. Future studies incorporating longitudinal immune profiling and molecular tumor characteristics may help clarify the relative contributions of host and tumor factors to survival outcomes in osteosarcoma.

While an elevated PLR was significantly associated with poorer outcomes in dichotomized univariable analysis, its prognostic value was attenuated and lost statistical significance upon multivariable adjustment. These findings are consistent with previous studies that linked elevated PLR and low LMR to adverse outcomes.^14,16,32 This correlation likely reflects an imbalance between host immune response and tumor-promoting inflammation, one of the ten hallmarks in cancer biology. PLR may signify a pro-tumor inflammatory state, with increased platelet-driven angiogenesis and immune suppression, while low LMR could indicate diminished lymphocyte-mediated antitumor activity and elevated monocyte levels that support tumor progression.^11,33 Unlike other published studies, the LMR in this study serves as a superior predictor compared to NLR or PLR. Several factors may explain this observation. First, since we chose the post-neoadjuvant, preoperative window rather than the pre-treatment setting, this would account for significant variations in immune cell counts and thus their respective ratios.³⁴ Second, the Vietnamese population may have a different genetic background and immune cell components compared to other cohorts.^35,36 In contrast, NLR did not demonstrate significant associations with survival outcomes in either univariable or multivariable models. Although NLR has been reported as a prognostic factor in several cancers, including sarcomas, our findings suggest that it may not be a reliable standalone marker in the osteosarcoma setting, particularly when using a data-derived cutoff of 6.57, which differs from previous studies.^13,37,38

With the focus on short-term survival outcomes, most progression events were observed to occur between 12 and 24 months postoperatively. The 24-month timepoint was selected because it is widely reported for progression-free survival and early outcome analyses, providing a relevant measure of short-term treatment efficacy.^39,40,41 Additionally, many relapses and disease progressions in osteosarcoma occur within the first two years after diagnosis, making the 24-month mark a meaningful indicator of early treatment success or failure and a relevant measure of short-term treatment efficacy.^41,42 In real-world cohorts, particularly in low- and middle-income countries such as Vietnam, longer follow-up is often limited by patient attrition, loss to follow-up, or resource constraints. Therefore, the 24-month cutoff also allows for a robust analysis while minimizing missing data. In the 24-month progression-free survival analysis, none of the examined biomarkers, including NLR, LMR, or PLR, demonstrated significant associations with outcomes. This lack of predictive value may reflect the limited discriminative ability of these markers, as indicated by AUC values close to 0.5.

Hgb also did not show a statistically significant association with survival. This finding aligns with a population-based nationwide study of 1,313 patients with esophageal cancer in Finland, in which preoperative Hgb levels had no independent prognostic value in esophageal cancer.⁴³ Interestingly, although 72% of patients presented with preoperative anemia, Hgb levels were not significantly associated with OS or PFS. Contrary to expectations, patients who experienced death had higher median Hgb levels compared to the overall cohort, with a median Hgb of 118.00 g/L versus 110.00 g/L, respectively. This counterintuitive trend may be due to clinical confounding, such as perioperative transfusions or pre-treatment optimization in more severely ill patients, which could mask the true prognostic relationship.

To further evaluate the robustness of our findings, we performed sensitivity analyses by modeling the inflammatory indices as continuous variables. The consistency of results across both dichotomized and continuous models is itself a meaningful finding, particularly with LMR retaining both the direction and approximate magnitude of its association with overall survival across the two modeling strategies. This observation suggests that the prognostic signal identified in the primary analysis is unlikely to be solely attributable to cutpoint selection. In addition, bootstrap internal validation demonstrated some reduction in discriminative performance after optimism correction, indicating potential model optimism that is expected in exploratory analyses based on relatively limited retrospective cohorts. Nevertheless, the direction and magnitude of the bootstrap-derived estimates remained broadly concordant with the primary analysis, supporting the overall stability of the observed associations.

In the continuous univariable analysis, both LMR and PLR maintained their prognostic significance, with LMR exhibiting a substantial protective effect (HR = 0.49, p = 0.049). LMR also achieved the highest discriminative accuracy among all tested indices (C-index = 0.729), outperforming NLR (C-index = 0.519) and PLR (C-index = 0.606). The shift in p values observed in the multivariable model (p = 0.085 for LMR and p = 0.068 for PLR) likely reflects the statistical attenuation common in smaller cohorts when multiple continuous covariates, such as age and gender, are adjusted simultaneously. From a biological perspective, the performance of LMR as a continuous variable reinforces our hypothesis that the post-neoadjuvant window serves as a critical stress test of the host. A higher lymphocyte-to-monocyte ratio during this interval likely represents superior host immune recovery and resilience following the myelosuppressive insult of chemotherapy. This systemic rebound may be as vital for long-term anti-tumor surveillance as the initial tumor burden itself.

Given the limited number of clinical events in our cohorts, making it susceptible to overfitting, we performed bootstrap internal validation with 1000 resamples to evaluate model stability. The multivariable model yielded an initial apparent C-index of 0.729. After correction, the apparent C-index of 0.729 fell to an optimism-corrected 0.628 (estimated optimism of 0.101), confirming a moderate degree of overfitting and suggesting that the model’s real-world discrimination is modest. Therefore, the overall clinical utility of the model must be stated with caution until it can be optimized and externally validated. Nevertheless, the bootstrap confidence intervals exposed distinct differences in parameter stability between the individual biomarkers. The bootstrap 95% CI for LMR narrowed and shifted entirely below the null threshold (0.188–0.776), demonstrating that its protective association remains stable despite sample size constraints. In contrast, the bootstrap CI for PLR crossed 1.0 (0.998–1.010), revealing that its prognostic value is unstable in this cohort and should be interpreted with caution.

Furthermore, while our cohort demonstrated a unique demographic profile, with a 62% female proportion and a predominantly pediatric/adolescent distribution, the stability of the LMR prognostic trend across both categorical and continuous modeling underscores its potential as a versatile tool. The exceptionally wide bootstrap CI observed for male gender (0.17–7.99) is a statistical consequence of a low event-per-variable (EPV) ratio across demographic subgroups, rather than instability in the biomarker dynamics themselves. When viewed alongside the 94% pathological homogeneity of our conventional osteosarcoma cases, these validation findings suggest that continuous LMR represents a robust, stable indicator of survival in this institutional setting, independent of specific thresholds used for clinical stratification.

To evaluate the generalizability of our findings, we applied the optimal cutoff values for NLR, LMR, and PLR reported in previous studies to our cohort to examine whether similar prognostic patterns emerged. Using the keywords “osteosarcoma”, “hematological biomarkers”, “inflammatory prognostic score”, “NLR”, “LMR”, “PLR”, and “prognosis”, we found 16 studies that enrolled a total of 2762 patients with osteosarcoma, mostly conducted in China. We observed that while the overall prognostic trends were often consistent with lower LMR and higher NLR generally correlating with poorer outcomes, the specific statistical significance frequently diminished when using external thresholds. Although several studies demonstrated that inflammatory markers were significantly associated with survival in univariable analysis, their significance often diminished in multivariable models. Some studies identified NLR and PLR as independent prognostic factors, while others reported significance only for LMR or found no independent biomarker. This inconsistency suggests these markers may be context-dependent and influenced by differences in confounder adjustment. As shown in Table 8, the optimal cutoff values for NLR, LMR, and PLR varied considerably across studies, ranging from 0.8 to 3.43 for NLR, 3.43 to 5.3 for LMR, and 115 to 200 for PLR. This heterogeneity likely reflects differences in study populations, statistical methods, and timing of blood collection. Across 16 studies, none of the biomarkers achieved an AUC of 0.8. AUC values exceeded 0.7 in 10 studies for NLR, but in none for LMR and only once for PLR, suggesting that NLR has generally demonstrated the strongest discriminative ability among the three. However, the overall modest AUC values indicate that while these markers may aid in risk stratification, they are unlikely to serve as reliable standalone prognostic indicators.

Table 8.

Previous study on prognostic value of NLR, LMR and PLR on osteosarcoma

No.	Author (Year)	NLR	LMR	PLR	Population (n, Country)
1	Liu et al.44	–	3.43 AUC 0.665	–	n = 327, China
2	Xia et al.45	3.43 AUC = 0.705	–	122 AUC = 0.606	n = 359, China
3	Liu et al.16	2.57 AUC = 0.663	4.73 AUC = 0.407	123.50 AUC = 0.600	n = 162, China
4	Vasquez et al.29	2.00, AUC = N/A	–	–	n = 55, Peru
5	Wang et al.46	1.00 AUC = 0.793	–	–	n = 290, China
6	Li et al.47	2.65 AUC = 0.644	–	118 AUC = 0.609	n = 216, China
7	Yang et al.14	2.65 AUC = 0.761	5.16 AUC = 0.683	125 AUC = 0.697	n = 77, China
8	Yang et al.48	2.96 AUC = 0.709	4.44 AUC = 0.640	–	n = 133, China
9	Yapar et al.32	2.45 AUC = 0.763	4.47 AUC = 0.603	119 AUC = 0.681	n = 172, Turkey
10	Ouyang et al.49	0.8 AUC = 0.626	–	–	n = 86, China
11	Tang et al.37	2.36 AUC = 0.793	–	115 AUC = 0.659	N = 96, China
12	Tian et al.50	2.53 AUC = 0.713	–	–	n = 87, China
13	Huang et al.38	2.04 AUC = 0.704	–	154.11 AUC = 0.665	n = 235, China
14	Li et al.51	2.77 AUC = N/A	–	–	n = 251, China
15	Erdogan et al.52	3.00 AUC = 0.773	5.3 AUC = 0.613	200 AUC = 0.734	n = 52, Turkey
16	Bastard et al.13	1.51 AUC = 0.93	–	–	n = 164, France

The median cutoff values reported across studies for NLR, LMR, and PLR were 2.55, 4.6, and 122.75, respectively. These reference values were applied to our cohort for survival analysis of OS and 24-month PFS, as shown in Figure 7. However, none of these cutoffs yielded statistically significant differences in survival, with all log rank p values greater than 0.05.

Figure 7.

OS and 24-month PFS survival analysis using reference cutoffs

(A) OS survival analysis for NLR.

(B) 24-month PFS survival analysis for NLR.

(C) OS survival analysis for LMR.

(D) 24-month PFS survival analysis for LMR.

(E) OS survival analysis for PLR; (F) 24-month PFS survival analysis for PLR.

Overall, our study contributes to the growing body of literature on systemic inflammatory markers. Our study demonstrates that the post-neoadjuvant, preoperative LMR is an independent predictor of overall survival in patients with osteosarcoma. LMR selectively predicted long-term OS but not 24-month early progression, suggesting LMR serves as a biological stress test of host systemic immune reconstitution and physiological resilience rather than a reflection of intrinsic tumor aggressiveness alone. The multi-layered validation performed in this osteosarcoma-focused study supports the stability of LMR over PLR and NLR, underscoring its clinical value as a simple, accessible biomarker for risk stratification in resource-limited settings such as Vietnam. However, the inconsistent performance of these systemic inflammatory indices across the literature limits confidence in their universal generalizability as standalone prognostic indicators in osteosarcoma, highlighting the need for cautious interpretation and further longitudinal prospective validation.

Limitations of the study

Nevertheless, our study has several limitations. First, the retrospective design and relatively small sample size from a single center reduce the statistical power and limit the generalizability of our findings to other populations. The small number of documented deaths during follow-up inherently limits the statistical power and parameter stability of multivariable survival models.

Second, many of our patients were transferred from other regional facilities with incomplete external records. This resulted in missing data for several established clinical prognostic factors, such as precise tumor site, size, surgical margins, and histological necrosis rate, restricting our ability to perform more extensive multivariable adjustments. While our predictive model focuses on peripheral blood ratios, large-scale registry data from the Cooperative Osteosarcoma Study Group have firmly established that conventional variables, such as axial tumor site, large tumor size, incomplete surgical remission, and poor histologic response to neoadjuvant chemotherapy, remain independent pillars of long-term survival.⁵³ This is further underscored by modern prospective data from the intercontinental EURAMOS-1 trial (n = 2260), which identified baseline pulmonary metastases and axial skeletal involvement as major adverse drivers, alongside real-world retrospective evidence confirming that surgical margin status and tumor necrosis rates below 90% significantly impair survival outcomes.^3,54

Third, peripheral biomarkers were assessed at a single pre-operative time point. The absence of longitudinal inflammatory biomarker measurements throughout the entire treatment course precluded the assessment of dynamic temporal changes and their potential prognostic implications before, during, and after neoadjuvant chemotherapy. Furthermore, while our protocol strictly isolated data collection within this specific post-neoadjuvant, pre-operative window, subtle variations in individual bone marrow recovery timing and the precise interval between the final chemotherapy cycle and blood sampling represent an unmeasured source of variance. Cytotoxic neoadjuvant chemotherapy induces profound, time-dependent myelosuppression, and although our sampling captured the immediate pre-operative baseline, future prospective protocols must strictly log chemotherapy-to-sampling intervals to isolate these effects.

Finally, we did not explicitly document baseline chronic inflammatory or autoimmune conditions, and data on concurrent systemic inflammatory status, episodes of active infection, recent blood transfusion, and nutritional metrics were not uniformly available across this historical dataset. Consequently, we cannot completely rule out residual confounding from subclinical systemic inflammation, intercurrent infection, blood transfusion, or nutritional variance, all of which are established physiological modulators of peripheral leukocyte and monocyte fractions. In adult populations, such comorbidities and associated oxidative stress can independently skew systemic inflammatory indices. However, given that the majority of our patients were children and adolescents (72%), the impact of chronic metabolic or systemic inflammatory diseases, which are more prevalent in older populations, is likely minimal. Future prospective studies are encouraged to incorporate a comprehensive baseline assessment of chronic inflammatory markers, nutritional co-variables, and relevant infection episodes, and to follow up with longitudinal trajectories to further isolate the specific impact of cancer biology on these peripheral blood ratios.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Son Hai Vu (son.vh@vinuni.edu.vn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

The clinical and laboratory datasets generated during this study are available from the corresponding author upon reasonable request. This paper does not report original code. Standard R packages used for statistical analysis are listed in the key resources table. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank Minh-Quang Dang and Ngoc-Minh Truong for critical discussion and assistance regarding data analysis. We also thank medical staff at the Vinmec Sarcoma Center for data curation and beyond. This study is funded by VinUniversity, Seed grant no. VUNI.2425.EME.007, to S.H.V.

Author contributions

Conceptualization: T.T.T.H., X.-H.N., and S.H.V. Investigation: T.T.T.H. and D.N.Q. Formal analysis: T.T.T.H. Validation: T.D.T., N.T.Q.S., T.T.T.H., and T.T.D. Writing – original draft: T.T.T.H., and S.H.V. Writing – review and editing: T.T.T.H. and S.H.V. Project administration: X.-H.N. and S.H.V. Methodology: T.T.T.H., D.N.Q., T.D.T., N.T.Q.S.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Clinical and laboratory dataset	This study	Available from Corresponding Author upon reasonable request
Software and algorithms
Jamovi (v2.6.25)	The Jamovi Project	RRID:SCR_016142
R (v4.5.2)	R Foundation for Statistical Computing	https://www.R-project.org/
survival R package (v3.8-6)	Therneau	https://CRAN.R-project.org/package=survival
rms R package (v8.1-1)	Harrell	https://CRAN.R-project.org/package=rms
Hmisc R package (v5.2-5)	Harrell	https://CRAN.R-project.org/package=Hmisc
Other
Digital Medical Records/Hospital Information System	Vinmec Healthcare System	N/A

Experimental model and study participant details

Human subjects

This retrospective observational cohort study evaluated 100 patients with a definitive pathological diagnosis of localized osteosarcoma (Enneking stage IB–IIB) who underwent their primary tumor resection at Vinmec Healthcare System, Vietnam, between January 2022 and March 2024.

Sex and gender influence

The association of sex with overall survival and progression-free survival was evaluated as a covariate in our multivariable Cox proportional hazards models, but no statistically significant influence was observed.

Sample allocation

This is a retrospective observational study of 100 human subjects, and patients were not randomized or actively allocated to experimental groups. Subjects were retrospectively stratified into groups based on their clinical outcomes (e.g., early progression vs. progression-free) for comparative statistical analyses.

Inclusion and exclusion criteria

Patients were included if they met the following criteria:

• 1.Histologically confirmed osteosarcoma (stage IB–IIB).
• 2.Received primary osteosarcoma-related surgery at Vinmec Healthcare System within the specified time frame.
• 3.Available preoperative complete blood count (CBC) records in the electronic hospital information system.
• 4.Completed a minimum post-surgical follow-up of 12 months (unless death occurred prior to 12 months).

Patients were excluded if they presented with:

• 1.Incomplete preoperative laboratory data.
• 2.Anti-cancer therapies administered prior to evaluation at Vinmec Healthcare System (excluding standard chemotherapy).
• 3.Administration of anti-inflammatory medications within one month prior to blood collection that could distort systemic inflammatory markers.

All surgical candidates demonstrated adequate renal function prior to surgery and chemotherapy administration, as defined by institutional clinical protocols (absence of severe renal impairment and laboratory parameters within acceptable limits for perioperative care). Age at diagnosis was dichotomized at 18 years to align with pediatric versus adult treatment frameworks.

Ethical approval

The protocol was reviewed and approved by the Institutional Ethical Review Board for Biomedical Research at Vinmec International General Hospital JSC – VinUniversity (Approval No. 111/CN/HDDD_VMEC). Informed consent was managed in accordance with institutional guidelines for retrospective electronic record reviews.

Method details

Data collection and preoperative sample processing

Demographic, clinical, pathological, and laboratory parameters were systematically extracted from the hospital information system. Variable categories included age, sex, age at diagnosis, Enneking stage, primary tumor location, presence of pathological fracture, metastatic status, and administration of neoadjuvant or adjuvant therapies.

Preoperative peripheral blood samples were drawn during standard pre-surgical assessments within one month prior to surgery. Samples were collected when patients were clinically stable and free of active infection or acute inflammatory conditions. In accordance with institutional protocols, all patients underwent definitive surgery within 3 weeks following their final cycle of neoadjuvant chemotherapy.

Biomarker calculations and cutoffs

Systemic inflammatory indices and hematological parameters were calculated from preoperative CBC data using the following formulas.

• •NLR: Absolute Neutrophil Count/Absolute Lymphocyte Count
• •PLR: Absolute Platelet Count/Absolute Lymphocyte Count
• •LMR: Absolute Lymphocyte Count/Absolute Monocyte Count

Anemia was defined according to age- and sex-adjusted threshold criteria: Hemoglobin (Hgb) <12 g/dL in adult non-pregnant females and children aged 12–14 years; <13 g/dL in adult males; and <11.5 g/dL in children aged 5–11 years.

Quantification and statistical analysis

Survival outcomes definition

Overall Survival (OS) was measured from the date of primary surgery to death from any cause. Progression-Free Survival (PFS) was measured from the date of primary surgery to the first recorded occurrence of disease progression, local recurrence, or death from any cause. Patients without an event were censored at the date of their last clinical follow-up. Early progression was further categorized by separating patients into “Progression within 24 months” versus “Progression-free ≥24 months”.

Statistical testing and modeling

Statistical analyses were executed using Jamovi (v2.6.25; RRID:SCR_016142) and R (v4.5.2) utilizing the survival (v3.8-6), rms (v8.1-1), and Hmisc (v5.2-5) packages.

Normality and descriptive testing

The normality of continuous variables (Hgb, NLR, LMR, and PLR) was evaluated using the Shapiro-Wilk test. Between-group comparisons for non-normally distributed continuous variables (e.g., biomarker differences between early progression vs. progression-free groups at 24 months) were conducted using the non-parametric Mann-Whitney U test.

ROC curve analysis

Receiver Operating Characteristic (ROC) curves were constructed to determine the discriminative ability of each biomarker and establish optimal cutoff values. Area Under the Curve (AUC) performance was categorized as failed (0.5–0.6), poor (0.6–0.7), fair (0.7–0.8), good (0.8–0.9), or excellent (0.9–1.0).

Survival analysis

Survival distributions between groups dichotomized by biomarker cutoffs were estimated using Kaplan-Meier curves and compared via the log rank test. Independent prognostic factors for OS and PFS were identified using univariate and multivariate Cox proportional hazards regression models. Model assumptions—including independence of observations, non-informative censoring, and proportional hazards—were formally evaluated and met. Hazard Ratios (HR) are reported alongside 95% Confidence Intervals (CI) and exact two-sided p-values.

Sensitivity and internal validation analyses

To address potential classification bias from outcome-derived threshold cutoffs and account for low OS event rates, sensitivity analyses were performed by modeling inflammatory indices (LMR, NLR, PLR) as continuous variables in Cox proportional hazards models. Proportional hazards assumptions were confirmed via Schoenfeld residuals. Model discrimination was measured using Harrell’s Concordance index (C-index). Internal validation and optimism correction were performed using bootstrapping with 1,000 resamples to assess parameter stability and generalizability, generating optimism-corrected C-indices and bootstrapped 95% CIs for hazard ratios.