This multicenter retrospective study was conducted at Nanfang Hospital Affiliated with Southern Medical University, Guangzhou Eighth People’s Hospital, Fourth Hospital of Nanning, and Chest Hospital of Guangxi Zhuang Autonomous Region in China. We enrolled 273 patients with initial AIDS-related DLBCL from January 2011 to December 2019, based on the 2008 WHO classification (23). The exclusion criteria consisted of patients with transformed DLBCL, primary central nervous system DLBCL, primary cutaneous DLBCL, and patients lacking complete physical examination, detailed disease history, treatment records, results, and follow-up data. A total of 153 patients (90 in the training cohort and 63 in the validation cohort) met the eligibility criteria ( Figure 1A ). The training cohort included patients from Nanfang Hospital and Chest Hospital of Guangxi Zhuang Autonomous Region, the other two hospitals contributed to the external validation cohort. Detailed inclusion and exclusion criteria are provided in the Supplementary Methods . The study obtained approval from the Ethics Committees of Nanfang Hospital (NFEC-2021-178), Guangzhou Eighth People’s Hospital (202210222), the Fourth People’s Hospital of Nanning ([2019]39), and the Chest Hospital of Guangxi Zhuang Autonomous Region (2022-022). Written informed consent was obtained from all participants in accordance with the World Medical Association Declaration of Helsinki.

Baseline characteristics, including age, sex, lactate dehydrogenase (LDH) levels, Eastern Cooperative Oncology Group performance status (ECOG PS), Ann Arbor stage, extranodal sites, Hb/RDW ratio, presence of B-symptoms, bulky disease, cell of origin, prior history of HIV, cART, CD4 levels, and the CD4/CD8 ratio, were collected. The primary endpoint of this study was OS, defined as the time from diagnosis to death from any cause.

The Hb/RDW ratio was integrated into the IPI score, and the incremental prognostic value was evaluated by comparing the area under the curve (AUC) of the receiver operating characteristic (ROC) curve between the IPI and the Hb/RDW-IPI risk score.

The optimal cutoff values for the biomarkers were determined using ROC curve analysis and the Youden index. Wilcoxon rank-sum tests or one-way analysis of variance on ranks were employed to compare continuous variables. Pearson’s and nonparametric Spearman’s correlations were used to assess the correlation between continuous variables. The Kaplan–Meier method was used to visualize the time-to-event variables, and log-rank tests were used to compare survival between cohorts. Univariate analysis identified potential prognostic variables correlated with OS. Variables significant at P < 0.05 in univariate analyses were included in the multivariable analysis using the Cox proportional hazards model. The association between variables and survival was summarized by the Hazard ratio (HR) and 95% confidence intervals (CI). Linear correlation analysis and the Student’s t-test were used to assess the relationship of the Hb/RDW ratio with other factors. Pearson’s chi-square was used to analyze the distributions of clinical characteristics between different groups. SPSS v.25.0 was used for all statistical analyses. All tests were two-sided, and significance was defined as a P < 0.05.

Between January 2011 and December 2019, four centers in southern China registered 273 cases of AIDS-related DLBCL, with 153 confirmed eligible ( Figure 1A ). A comparison between the training cohort (n = 90) and the validation cohort (n = 63) revealed no significant differences in demographics and other parameters ( Table 1 ). The median duration of follow-up was 12.0 months (95% CI, 8.5–15.5; range, 1 to 104 months), and the OS rate was 35.9% ( Figure 1B , Supplementary Figure 1 ). Specifically, patients with AR-DLBCL exhibited 1-, 2-, and 3-year OS rates of 49.7%, 39.9%, and 37.3%, respectively. The median age was 47 years [interquartile range (IQR) 39–58], with 128 patients (83.7%) being male. Among them, 31 individuals (20.3%) were over 60 years old, 118 (77.1%) had LDH > upper limit of normal (ULN), 76 (49.7%) had ECOG PS > 1, 78 (51.0%) had Ann Arbor stage III/IV disease, 44 (28.8%) had extranodal disease at > 1 site ( Table 1 ), and 62 (40.6%) had IPI high-intermediate–risk and high–risk scores ( Supplementary Table 1 ). The median Hb/RDW ratio was 8.4 (IQR 6.0–10.1). The optimal cutoff for the Hb/RDW ratio, determined through ROC curve analysis ( Supplementary Figure 3A ), was 8.5 (AUC = 0.737, 95% CI = 0.658–0.816, P < 0.001). Among the patients, 80 (52.3%) had a Hb/RDW ratio < 8.5. B symptoms were found in 25 patients (16.3%), bulky disease (> 7.5 cm) was documented in 57 individuals (37.3%), and the germinal center B-cell (GCB) subtype was observed in 81 patients (52.9%). Before the onset of AR-DLBCL, 38 out of 153 patients (24.8%) had a history of HIV infection. Among them, 29 (76.3%) were on cART for over 3 months, and 5 (13.2%) lacked treatment records. Of those on cART, 9 out of 29 (31.0%) had an undetectable HIV-1 viral load, while 9 (31.0%) had unknown viral load levels, partly due to incomplete historical records. Additionally, 97 patients (63.4%) had received cART at the baseline of lymphoma treatment. The median CD4+ T cell count and the CD4/CD8 ratio were 171 cells/µL (IQR 73–274) and 0.26 (IQR 0.15–0.47), respectively ( Table 1 ).

In the training cohort, univariate analysis revealed that age, LDH, Hb/RDW ratio, CD4 count, and the CD4/CD8 ratio were predictive of OS ( Table 2 ). In multivariable analysis, considering DLBCL, HIV, and inflammation-related factors, age (HR = 2.873, 95% CI = 1.522–5.421, P = 0.001), LDH levels (HR = 2.421, 95% CI = 1.058–5.537, P = 0.036), the Hb/RDW ratio (HR = 2.645, 95% CI = 1.267–5.522, P = 0.010), and the CD4/CD8 ratio (HR = 2.151, 95% CI = 1.076–4.301, P = 0.030) remained independent predictors of clinical outcome ( Table 2 , Supplementary Figure 2 ). In particular, only the Hb/RDW ratio (HR = 2.645, 95% CI = 1.267–5.522, P = 0.010) retained its independent prognostic value in the external validation cohort ( Table 2 ).

Kaplan–Meier analysis highlighted a significant association between a low Hb/RDW ratio and poor outcomes ( Figures 2A, B ). Similarly, this novel inflammatory marker, the Hb/RDW ratio (HR = 2.022, 95% CI = 1.316–3.106, P = 0.001), consistently demonstrated prognostic significance in the overall cohort ( Figure 2C ). An lower Hb/RDW ratio was associated with adverse clinical factors ( Figure 3A ), including the advanced Ann Arbor stage (R^2 = 0.035, P = 0.021), increased extranodal sites (R^2 = 0.054, P = 0.004), reduced CD4 count (R^2 = 0.039, P = 0.014), elevated levels of LDH (P = 0.012), poorer ECOG PS (P = 0.001), and higher IPI score (P = 0.010). The chi-square test ( Table 3 ) demonstrated a significant correlation between the Hb/RDW ratio and LDH levels (P = 0.015), Ann Arbor stage (P = 0.044), and the extranodal sites (P = 0.012).

Furthermore, we conducted additional analyses to evaluate the significance of the Hb/RDW ratio within different subgroups based on the LDH, ECOG PS, and IPI scores ( Figures 3B, C ). Among the AR-DLBCL population, 32.4%, 50.7%, and 68.0% exhibited a low Hb/RDW ratio in groups with LDH levels < 250, 250-750, and > 750 U/L, respectively (P = 0.012). Similarly, among patients with ECOG PS 0, 1, 2, 3, and 4, the proportion of patients with a reduced Hb/RDW ratio were 40.0%, 46.0%, 56.1%, 64.3%, and 100%, respectively (P = 0.001). A comparable trend was observed in the IPI score groups, with the percentages of individuals having a low Hb/RDW ratio at 37.2%, 46.5%, 67.7%, and 71.0% in the IPI score 1, 2, 3, and 4 groups, respectively (P = 0.010).

Incorporating the Hb/RDW ratio into the IPI score defined the Hb/RDW-IPI scores, with each variable scored as 1 based on the hazard ratio of each category in the total cohort ( Figure 2C , 16), which allowed us to significantly improve survival prediction ( Figure 4 ). The entire cohort was accurately classified into three risk groups: low-risk (scores: 0–1; 17.6%), intermediate-risk (scores: 2–4; 67.3%), and high-risk (scores: 5–6; 15.0%), corresponding to OS rates of 74.1%, 32.0%, and 8.7%, respectively. The combined Hb/RDW-IPI demonstrated a statistically significant stratification of the three risk groups (P < 0.001) and exhibited superior distinction for the intermediate- and high-risk groups compared to the IPI score alone. Hb/RDW-IPI showed a moderate improvement in discrimination compared to the original IPI alone ( Supplementary Figure 3B , AUC [95% CI]: Hb/RDW-IPI, 0.70 [0.61–0.78] vs. IPI, 0.67 [0.58–0.77]).