A predictive model for HIV-related lymphoma

Abstract

Objectives:

To address the paucity of HIV-related lymphoma (HRL)–specific prognostic scores for the Japanese population by analyzing domestic cases of HRL and constructing a predictive model.

Design:

A single-center retrospective study coupled with a review of case reports of HRL.

Methods:

We reviewed all patients with HRL treated at our hospital between 2007 and 2023 and conducted a comprehensive search for case reports of HRL from Japan using public databases. A multivariate analysis for overall survival (OS) was performed using clinical parameters, leading to the formulation of the HIV-Japanese Prognostic Index (HIV-JPI).

Results:

A total of 19 patients with HRL were identified in our institution, whereas the literature review yielded 44 cases. In the HIV-JPI, a weighted score of 1 was assigned to the following factors: age at least 45 years, HIV-RNA at least 8.0×10⁴ copies/ml, Epstein–Barr virus-encoded small RNA positivity, and Ann Arbor classification stage IV. The overall score ranged from 0 to 4. We defined the low-risk group as scores ranging from 0 to 2 and the high-risk group as scores ranging from 3 to 4. The 3-year OS probability of the high-risk group [30.8%; 95% confidence interval (CI): 9.5–55.4%) was significantly poorer than that of the low-risk group (76.8%; 95% CI: 52.8–89.7%; P < 0.01).

Conclusion:

This retrospective analysis established pivotal prognostic factors for HRL in Japanese patients. The HIV-JPI, derived exclusively from Japanese patients, highlights the potential for stratified treatments and emphasizes the need for broader studies to further refine this clinical prediction model.

Introduction

Even in the era of highly active antiretroviral therapy (ART), HIV-related lymphoma (HRL) remains one of the most common HIV-associated cancers and the leading cause of HIV-related deaths [1–8]. Numerous studies have documented the unique clinical presentations of HRL, highlighting their divergence from non-HRLs [9–24]. However, there remains a scarcity of prognostic models specifically designed for HRL, underscoring a critical need for tailored clinical prognostication [25–27].

To address this knowledge gap, the present study rigorously analyzed a cohort of HRL cases managed at our institution and reviewed relevant domestic case reports to construct a predictive model that aims to refine prognostic assessments for this distinct patient group.

Methods

Case series and literature review

The study was approved by the Ethics Committee of Yokohama Municipal Citizen's Hospital (designated approval number: 23-04-03) and was performed in accordance with the principles of the Declaration of Helsinki. All participants provided informed consent for the use of their data.

A flowchart of the retrospective analysis is shown in Fig. 1. We reviewed all patients with HRL treated at Yokohama Municipal Citizen's Hospital between 2007 and 2023 using the clinical laboratory database and medical records (YMCH cohort). For the scope of HRL, our inclusion criteria focused on patients with a documented history of HIV infection who were subsequently diagnosed with B-cell lymphoma, confirmed through histopathological examination. The following data were retrieved: patients’ age, sex, year of lymphoma diagnosis, duration from the diagnosis of HIV infection to the identification of lymphoma, CD4⁺ counts and HIV-ribonucleic acid (RNA) PCR levels at diagnosis, ART before and after HRL diagnosis, coexisting AIDS-defining illnesses, pathological diagnosis, biopsy site, Ann Arbor classification, extranodal sites, Epstein–Barr virus-encoded small RNA (EBER) status, treatment regimen, and outcomes. AIDS-defining illnesses were identified according to the 1993 classification from the Centers for Disease Control and Prevention [28].

Fig. 1.

Flowchart of patient selection and study design.

ALCL, anaplastic large cell lymphoma; HL, Hodgkin lymphoma; HRL, HIV-related lymphoma; N, number of patients; PBL, plasmablastic lymphoma; PEL, primary effusion lymphoma; T/NKCL, T/NK-cell lymphoma; YMCH, Yokohama Municipal Citizen's Hospital.

To further scrutinize the clinical characteristics of HRL in Japan, we undertook an extensive search for Japanese case reports published between 2007 and 2023, documented in either Japanese or English, using PubMed and Ichushi-Web (Japan Medical Abstracts Society). PubMed searches were conducted using the following keywords

‘AIDS-related lymphoma’, ‘HIV-related lymphoma’, ‘AIDS-associated lymphoma’, (‘lymphoma’ AND ‘human immunodeficiency virus’), (‘lymphoma’ AND ‘Acquired immunodeficiency syndrome’), (‘lymphoma’ AND ‘AIDS’), OR (‘lymphoma’ AND ‘HIV’) AND Japan. Taking into consideration the composition of HRL in Japan and the differences in clinical characteristics among the various subtypes [9–13], we excluded cases of T-cell/natural killer-cell lymphoma, Hodgkin lymphoma, primary effusion lymphoma, and plasmablastic lymphoma. Additionally, case reports lacking survival data were also omitted from the study. Clinical information akin to that collected for the YMCH cohort was gathered from the identified cases (literature cohort).

Statistical analysis

The primary endpoint of this study was to assess the overall survival (OS). The OS was defined as the time from the diagnosis of HRL to death or the time of last contact. The probability of OS was estimated using the Kaplan–Meier method. Univariate and multivariate analyses were performed using the Cox proportional hazard regression model for OS, and hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated. The final model variables were determined using Akaike information criterion (AIC) model selection [29,30]. The variables for this analysis were clinically determined based on previous reports [15–27]. Based upon the AIC-selected variables, we calculated the risk score for patients with HRL [HIV-Japanese Prognostic Index (HIV-JPI)]. Points were assigned to each factor, and a cumulative score was calculated for each case. Stratification into high-risk and low-risk groups was then performed based on the total scores, and OS comparisons were made between these strata. When comparing the OS, P values were calculated using the log-rank test. All tests were two-sided, and a P value of less than 0.05 was considered statistically significant. All statistical analyses were performed using EZR, a graphical user interface for R software (The R Foundation for Statistical Computing, version 4.3.0, Vienna, Austria) [31].

Handling of missing data

The procedure used for handling missing data has been described in a previous study [32]. We imputed missing values under the assumption of missing values at random using a nonparametric algorithm using the random forest method [33,34]. In this method, missing values were imputed using predictions from other variables. The performance of this method was comparable to that of parametric methods, such as multiple imputations by chained equations [35,36]. The imputation was performed using the ‘missforest’ package in the R software.

In our comprehensive approach to the challenge of missing data, we also devised a supplementary prognostic index (SPI) employing listwise deletion. Unlike the formulation of the HIV-JPI, we omitted any variables from the analysis that displayed over 10% data unavailability, ensuring the evaluation encompassed only cases with fully recorded data.

Results

Case series from our institution (Yokohama Municipal Citizen's Hospital cohort)

The clinical charts of 1120 patients with lymphoma at our institution were reviewed, and we identified 21 patients (1.8%) with HRL, including those in our previous case reports [37,38]. Among them, patients with Hodgkin lymphoma (n = 1) and those with anaplastic large cell lymphoma (n = 1) were excluded from the present study. In the cohort, 18 patients were of Japanese nationality, and one was French.

The clinical features of 19 patients are summarized in Table 1. At the time of HRL diagnosis, the median age of the patients was 50 years (range: 37–80 years). All participants were men. At the time of their diagnosis, median CD4⁺ counts and HIV-RNA levels were 115/μl (range: 1–743/μl) and 4.7×10⁴ copies/ml (range: <2.0×10²–5.1×10⁶ copies/ml), respectively. Thirteen patients (68.4%) had not commenced ART at the point of their HRL diagnosis. Following their diagnosis with HRL, all patients initiated ART. The initial ART regimens varied; some of the commonly used combinations included abacavir/lamivudine (ABC/3TC) and raltegravir (RAL; n = 3); tenofovir alafenamide/emtricitabine (TAF/FTC) and dolutegravir (DTG; n = 3); ABC/3TC and fosamprenavir (n = 2); FTC, stavudine, and nelfinavir (n = 2); TAF/FTC and RAL (n = 2). Five patients (26.3%) presented with AIDS-defining illnesses, including pneumocystis pneumonia (n = 3, 15.8%), candidiasis (n = 2, 10.5%), cytomegalovirus infection (n = 1, 5.3%), and tuberculosis (n = 1, 5.3%).

Table 1.

Patients’ clinical features in Yokohama Municipal Citizen's Hospital cohort.

	At the time of HRL diagnosis															After HRL diagnosis
No [Ref]	Age (years)	Sex	National origin	Year	HIV to Dx (months)	CD4+ cell count (cells/μl)	HIV-RNA (copies/ml)	ART	ADIs	LDH (U/l)	sIL-2R (U/ml)	Disease type	Biopsy site	CS	CNS	EBER	Treatment (number of cycles)	Initial ART	Outcome	Cause of death	Follow-up (months)
1 [37]	37	M	Japan	2007	2	35	NA	None	−	570	1653	BL	GI	IV	+	NA	EPOCH [2] hyper-CVAD [6]	FTC d4T NFV	Alive	−	198
2	37	M	Japan	2007	5	31	2.4 × 105	None	PCP NTM CMV	491	1174	DLBCL	Liver	IV	−	NA	R [1] EPOCH-R[1] Bleomycin [1]	TDF FTC EFV	Death	Lymphoma	3
3	62	M	Japan	2007	0	5	3.2 × 104	None	−	818	52 053	DLBCL	LN	IV	−	NA	No chemotherapy	FTC d4T NFV	Death	Lymphoma	0
4	62	M	Japan	2008	6	334	1.8 × 104	None	−	179	1280	DLBCL	LN	IV	+	−	EPOCH [6]	ABC/3TC FPV	Alive	−	179
5	56	M	Japan	2011	72	340	1.6 × 103	None	−	6875	NA	BL	LN	IV	+	−	hyper-CVAD [1] HD-MTX/AraC [1]	ABC/3TC FPV	Death	Lymphoma	3
6	37	M	Japan	2011	0	1	4.7 × 104	None	−	212	NA	DLBCL	Brain	IV	+	−	WBRT EPOCH[6]	ABC RAL DRV/r	Alive	−	143
7	59	M	Japan	2012	0	75	5.1 × 106	None	PCP Candida	338	4243	DLBCL	Skin	IV	−	+	R-CHOP [1] HD-MTX [1]	3TC RAL FPV	Death	Sepsis	3
8	50	M	Japan	2013	0	154	6.0 × 104	None	−	2,477	2277	BL	GI	IV	+	+	R-CHOP [1] R-hyper-CVAD[1] EPOCH-R [1]	ABC/3TC RAL	Death	Myocardial infarction	89
9	80	M	Japan	2013	0	184	2.2 × 105	None	−	126	210	DLBCL	Liver	IV	−	NA	R-CHOP [6]	ABC/3TC RAL	Alive	−	61
10	46	M	Japan	2014	23	743	BDL	TDF/FTC DRV rtv	Tb	557	NA	DLBCL	GI	IE	−	NA	R-CHOP [6]	TDF FTC RAL	Alive	−	113
11	49	M	Japan	2014	61	300	BDL	ABC/3TC RAL	−	288	7271	SMZL	BM	IV	−	+	EPOCH [2]-R [4] R-ESHAP [1] R-ICE [3], CBT	ABC/3TC RAL	Death	Encephalopathy	20
12	59	M	Japan	2018	154	86	BDL	TAF/FTC RAL	−	235	1177	DLBCL	LN	IV	−	+	EPOCH-R [6]	TAF/FTC RAL	Alive	−	58
13	43	M	Japan	2020	0	115	2.3 × 104	None	−	575	1594	DLBCL	LN	II	−	+	EPOCH-R [3]	TAF/FTC DTG	Death	Lymphoma	7
14	45	M	Japan	2020	150	137	BDL	BIC/TAF/FTC	−	325	3638	TCRLBCL	LN	IV	−	+	R-CHOP [6]	BIC/TAF/FTC	Alive	−	34
15	51	M	Japan	2020	1	77	4.6 × 105	None	PCP Candida	187	716	BL	GI	IE	−	−	EPOCH-R [6]	TAF/FTC DRV/cobi	Alive	−	34
16	51	M	Japan	2020	0	17	3.1 × 105	None	−	849	1783	BL	LN	IV	−	NA	EPOCH-R [3] HD-MTX/AraC [2] ASCT, WBRT	TAF/FTC DTG	Death	Lymphoma	10
17 [38]	44	M	Japan	2021	116	52	2.0 × 106	None	−	335	5466	DLBCL	LN	III	−	+	R-CHOP [6] ESHAP [1] PBR[4], CBT	TAF/FTC RAL	Alive	−	24
18	27	M	Japan	2022	15	583	BDL	TAF/FTC DTG	−	3161	873	BL	BM	IV	−	−	hyper-CVAD [4] HD-MTX/AraC [4]	TAF/FTC DTG	Alive	−	20
19	51	M	France	2022	82	409	BDL	DTG/3TC	−	192	4406	FL	LN	IV	−	−	BR [6]	DTG/3TC	Alive	−	12

3TC, lamivudine; ABC, abacavir; ADIs, acquired immunodeficiency syndrome defining illnesses; ART, antiretroviral therapy; ASCT, autologous stem cell transplantation; BDL, below detection limit; BIC, bictegravir; BL, Burkitt lymphoma; BM, bone marrow; BR, bendamustine/rituximab; CBT, cord blood transplantation; CNS, central nervous system; CMV, cytomegalovirus; cobi, cobicistat; CS, clinical stage; d4T, stavudine; DLBCL, diffuse large B-cell lymphoma; DRV, darunavir; DTG, dolutegravir; Dx, diagnosis; EBER, Epstein–Barr virus-encoded small RNA; EFV, efavirenz; EPOCH-R, etoposide/prednisolone/vincristine/cyclophosphamide/doxorubicin/rituximab; FL, follicular lymphoma; FPV, fosamprenavir; FTC, emtricitabine; GI, gastrointestinal tract; HD-MTX/AraC, high-dose methotrexate/cytarabine; LN, lymph node; M, male; NA, not available; NFV, nelfinavir; NTM, nontuberculous mycobacteria; PBR, polatuzumab vedotin with BR; PCP, pneumocystis pneumonia; R, rituximab; R-CHOP, rituximab/cyclophosphamide/doxorubicin/vincristine/prednisone; R-ESHAP, rituximab/etoposide/methylprednisolone/cytarabine/cisplatin; R-hyper-CVAD, rituximab/cyclophosphamide/vincristine/doxorubicin/dexamethasone; R-ICE, rituximab/ifosfamide/carboplatin/etoposide; RAL, raltegravir; r or rtv, ritonavir; sIL-2R, soluble interleukin-2 receptor; SMZL, splenic marginal zone lymphoma; TAF, tenofovir alafenamide; Tb, Tuberculosis; TDF, tenofovir disoproxil fumarate; TCRLBCL, T-cell/histiocyte-rich large B-cell lymphoma; WBRT, whole brain radiation therapy; YMCH, Yokohama Municipal Citizen's Hospital cohort.

Ten patients (52.6%) were diagnosed with diffuse large B-cell lymphoma, of whom 15 (79.0%) were diagnosed with stage IV disease. In addition, seven patients (53.8%) tested positive for EBER. For initial chemotherapy, nine patients (50.0%) received EPOCH (etoposide/prednisolone/vincristine/cyclophosphamide/doxorubicin) with or without rituximab, and six patients (31.5%) received CHOP (cyclophosphamide/doxorubicin/vincristine/prednisone) with or without rituximab. At a median follow-up period of 58 months (range: 12–198 months), 11 patients (57.9%) remained alive. Lymphoma was the most prevalent cause of death in these patients (n = 5); other causes included sepsis (n = 1), myocardial infarction (n = 1), and encephalopathy (n = 1).

Development of the HIV-Japanese Prognostic Index for patients with HIV-related lymphoma

To explore domestic prognostic factors of HRL in Japan, we combined data from 44 episodes of HRL identified in the literature cohort – comprising 18 single case reports and 2 case series published between 2007 and 2023 [39–58] – with our YMCH cohort. We then conducted a consolidated multivariate analysis of OS across both cohorts. The clinical features of the literature cohort are summarized in Supplementary Table 1. The final model variables determined by the AIC were age at least 45 years (hazard ratio: 2.02, 95% CI: 0.86–4.74, P = 0.11), HIV-RNA at least 8.0×10⁴ copies/ml (hazard ratio: 2.24, 95% CI: 0.96–5.24, P = 0.062), EBER positivity (hazard ratio: 2.36, 95% CI: 0.82–6.73, P = 0.11), and Ann Arbor classification stage IV (hazard ratio: 2.36, 95% CI: 0.87–6.38, P = 0.091; Table 2). In the univariate and multivariate analyses, imputed values generated by the missforest package were used in place of missing data [33,34].

Table 2.

Univariate and multivariate analyses for development of the HIV-Japanese Prognostic Index: overall survival.

			Overall survival
			Univariate analysis			Multivariate analysis
Factor	Group	N (%)	HR	(95% CI)	P	HR	(95% CI)	P	HIV-JPI score
Age	<45 years	29 (46.0)	Ref			Ref
	≥45 years	34 (54.0)	1.75	(0.78-3.94)	0.17	2.02	(0.86–4.74)	0.11	1
CD4+ cell count	≥80 cells/μl	30 (50.0)	Ref
	<80 cells/μl	30 (50.0)	1.40	(0.63–3.09)	0.41
HIV-RNA PCR	<8.0 × 104 copies/ml	29 (50.9)	Ref			Ref
	≥8.0 × 104 copies/ml	28 (49.1)	2.13	(0.92–4.93)	0.078	2.24	(0.96–5.24)	0.062	1
History of ART	Yes	18 (28.6)	Ref
	No	45 (71.4)	1.47	(0.59–3.66)	0.41
AIDS-defining illnesses	No	40 (72.7)	Ref
	Yes	15 (27.3)	1.23	(0.49–3.13)	0.66
Pathological diagnosis	FL, DLBCL, others	40 (69.0)	Ref
	BL, HGBL	18 (31.0)	0.74	(0.3–1.85)	0.52
EBER	Negative	21 (55.3)	Ref			Ref
	Positive	17 (44.7)	2.1	(0.79–5.54)	0.13	2.36	(0.82–6.73)	0.11	1
Ann Arbor classification	I, II, III	19 (30.2)	Ref			Ref
	IV	44 (69.8)	2.31	(0.87–6.13)	0.093	2.36	(0.87–6.38)	0.091	1
CNS involvement	No	46 (73.0)	Ref
	Yes	17 (27.0)	1.09	(0.47–2.51)	0.84

AIDS, acquired immune deficiency syndrome; ART, antiretroviral therapy; BL, Burkitt lymphoma; CI, confidence interval; CNS, central nervous system; DLBCL, diffuse large B-cell lymphoma; EBER, Epstein–Barr virus-encoded small RNA; FL, follicular lymphoma; HGBL, high-grade B-cell lymphoma; HR, hazard ratio; JPI, Japanese Prognostic Index, Ref, reference.

For each prognostic factor identified through multivariate analysis, we assigned 1 point for age at least 45 years, 1 point for HIV-RNA at least 8.0×10⁴ copies/ml, 1 point for EBER positivity, and 1 point for Ann Arbor classification stage IV. As shown in Supplementary Table 2, the total points were defined as the HIV-JPI score, with 0–2 points categorized as low risk and 3–4 points as high risk. Cases with missing values were excluded from the risk stratification. We established two risk categories using HIV-JPI scores: the low-risk group from 23 cases (YMCH cohort, 9 cases; literature cohort, 14 cases), and the high-risk group with a median score of 3 (range 3–4) from 13 cases (YMCH cohort, 4 cases; literature cohort, 9 cases) (Supplementary Table 3). The 3-year OS probability of the high-risk group (30.8%; 95% CI: 9.5–55.4%) was significantly poorer than that of the low-risk group (76.8%; 95% CI: 52.8–89.7%; P < 0.01; Fig. 2a). The area under the curve (AUC) value was 0.74 (95% CI: 0.57–0.91), indicating a moderate discriminatory ability of the model to accurately classify our cohort [59]. In the YMCH cohort, although there was no significant difference, the 3-year OS was poorer in the high-risk group compared with the low-risk group [50.0% (95% CI: 9.5–55.4%) vs. 77.8% (95% CI: 36.5–93.9%); P = 0.19; Fig. 2b]. In the literature cohort, the high-risk group demonstrated significantly poorer outcomes than the low-risk group [22.2% (95% CI: 3.4–51.3%) vs. 75.7% (95% CI: 41.3–91.6%); P < 0.01; Fig. 2c).

Fig. 2.

Overall survival of patients with HIV-related lymphoma according to the HIV-Japanese Prognostic Index.

(a) Whole cohort; (b) YMCH cohort; and (c) literature cohort. A weighted score of 1 was assigned to the following factors: age at least 45 years, HIV-RNA at least 8.0 × 10⁴ copies/ml, EBER positivity, and Ann Arbor classification stage IV. The total points were defined as the HIV-JPI score, with 0–2 points categorized as low risk and 3–4 points as high risk. P values were calculated using the log-rank test. CI, confidence interval; EBER, Epstein–Barr virus-encoded small RNA; N, number of patients; OS, overall survival; YMCH, Yokohama Municipal Citizen's Hospital.

To address the issue of missing data from multiple perspectives, we also developed a SPI utilizing listwise deletion, as presented in Supplementary Table 4. Differing from the development of the HIV-JPI, factors with more than 10% missing data, specifically EBER positivity (n = 25, 39.7%) and AIDS-defining illnesses (n = 8, 12.7%), were excluded from the analysis. We analyzed 51 cases (81.0%) that had complete data across variables. The final model variables were age at least 45 years (hazard ratio: 2.68, 95% CI: 0.97–7.42, P = 0.057) and Ann Arbor classification stage IV (hazard ratio 3.20, 95% CI: 0.92–11.10, P = 0.067). For each prognostic factor extracted from the multivariate analysis, we allocated 1 point for age at least 45 years and 1 point for Ann Arbor classification stage IV. The aggregate of these points constituted the SPI score, with 0 or 1 point designated as low risk and 2 points as high risk (Supplementary Table 5). The 3-year OS probability for the high-risk cohort (43.8%; 95% CI: 22.3–63.4%) was significantly lower than that for the low-risk cohort (78.3%; 95% CI: 57.8–89.6%; P < 0.01; Supplementary Fig. 1), with an AUC value of 0.70 (95% CI: 0.56–0.83).

Discussion

In this study, we meticulously examined domestic cases of HRL, culminating in the creation of a unique prognostic score for the HRL endemic in Japan. Supplementary Table 6 delineates the international comparison of the characteristics of patients with HRL from recent studies [18–22]. Compared with prior reports, a higher prevalence of male patients was found in this study presenting with markedly reduced CD4⁺ counts and elevated HIV-RNA levels at the diagnosis of HRL. Furthermore, advanced disease stages and the presence of extranodal lesions were predominant, similar to those reported in prior studies. The global AIDS pandemic and treatment strategies vary significantly among the countries owing to various factors, emphasizing the importance of regional approaches to HIV care [60]. Therefore, a prognostic score derived exclusively from Japanese patients provides invaluable insights specific to Japan, emphasizing the international diversity in the realm of HIV treatment.

In this study, we present a unique case series detailing ART regimens in patients with HRL, an area of research that remains underexplored. The antiretroviral drugs approved in Japan align with those sanctioned in the United States and Europe [61]. According to the Japanese guidelines, the recommended ART regimens for HRL are combinations with ABC/3TC and DTG, TDF/FTC and DTG, and TDF/FTC and RAL owing to their reduced potential for drug interactions [6]. Furthermore, Japan approved the use of TAF in 2016, a drug that presents fewer adverse effects, such as renal tubular disorders and decreased bone density, compared with TDF [62,63]. In this study, the ART regimen was determined based on the discretion of the attending physicians, guided by these reported findings. In the era of combination ART, the administration of ART during chemotherapy for HRL is supported by numerous studies [15,16,64–66]. In addition, ART regimens are undergoing rapid transformation, such as the advent of intramuscular formulations [67–69], which could be a promising option when oral intake is compromised because of the side effects of chemotherapy [38]. Thus, the optimal ART regimen in patients with HRL warrants further exploration and validation.

Table 3 presents a compilation of the present study juxtaposed with that of previous reports concerning the clinical predictive models for HRL [25–27]. In the current study, beyond the factors encompassed by the International Prognostic Index, EBER expression and HIV-RNA levels were selectively incorporated into the HIV-JPI [70]. The implications of EBV expression on the prognosis vary across reports [71–75]. However, correlations not only with the prognosis but also with CD4⁺ counts and the onset of HRL have been reported [71]. The quantity of HIV-RNA has been reported to be useful in predicting the prognosis of HRL [76–78]. Thus, these factors could be more important when considering the prognosis of HRL compared with that of non-HRL. In addition to these factors, the treatment of HRL encompasses a multitude of considerations by integrating factors related to HIV (such as CD4⁺ counts, AIDS-defining illnesses, and combined ART) as well as those associated with physician practices (such as center effects on HIV and lymphoma treatment) [7,8]. Such complexities underscore the significance of a prognostic prediction model tailored specifically for HRL, as highlighted in the present study.

Table 3.

Clinical prediction models for HIV-related lymphoma from recent studies.

Ref (model)	Country	Year	Disease	Prognostic factors	Score = risk classification	OS (95% CI)
Present study (HIV-JPI)	Japan	2023	DLBCL BL Others	Age ≥45 years HIV-RNA ≥8.0×104 copies/ml EBER positivity Ann Arbor classification stage IV	0–2 = low (n = 33) > 2 = high (n = 30)	At 3 years Low: 80.5% (61.5–90.8%) High: 39.4% (21.0–57.3%)
ARL-IPI [25]	United States	2014	DLBCL BL Others	LDH abnormal Ann Arbor classification stage III or IV ECOG performance status ≥2 Extranodal site HIV score incorporating CD4+, HIV viral load, and history of AIDS	0–6 = low (n = 41) 7–10 = intermediate (n = 85) 11–15 = high (n = 28)	At 5 years Low: 78% (46–93%) Intermediate: 60% (40–75%) High: 50% (23–72%)
HIV-IPI [26]	China	2022	DLBCL BL Others	Age >60 years LDH abnormal ECOG performance status ≥2 Ann Arbor classification stage III or IV Extranodal site ≥2 CD4/CD8 ratio < 0.41	0–1 = low (n = 30) 2 = low–intermediate (n = 31) 3 = high–intermediate (n = 33) 4–6 = high (n = 44)	NA
New model [27]	China	2023	DLBCL	Age >60 years LDH ratio >1 Ann Arbor classification stage III or IV Bulky disease Elevated RDW level	0–1 = low (n = 20) 2–3 = intermediate (n = 37) 4–5 = high (n = 12)	At 3 years Low: 75.0% (54.5–84.2%) Intermediate: 40.5% (31.0–58.0%) High: 0.0% (4.1–13.6%)

BL, Burkitt lymphoma; CI, confidence interval; DLBCL, diffuse large B-cell lymphoma; EBER, Epstein–Barr virus-encoded small RNA; ECOG, Eastern Cooperative Oncology Group; NA, not available; OS, overall survival; RDW, red cell distribution width; Ref, reference.

The YMCH and literature cohorts showed that initial chemotherapeutic regimens with EPOCH or CHOP, with or without rituximab, did not affect the OS (data not shown). A consensus on the gold-standard therapy for HRL has yet to be established [5,6]. A large-scale meta-analysis has indicated the superiority of EPOCH-R over R-CHOP in treating HIV-associated DLBCL [15]. However, no study performed a direct comparison between EPOCH-R and R-CHOP in this population [8]. The efficacy of CODOX-M/IVAC and EPOCH has been reported in the treatment of HIV-associated Burkitt lymphoma [79–81]. Unlike non-HRLs, the administration of rituximab is controversial owing to the elevated risk for opportunistic infections [82–84]. Although previous studies have reported that there is no association between the use of rituximab and the risk of infection-related death [84], it is recommended to maximize the prophylaxis for opportunistic infections in patients with CD4⁺ counts below 50 cells/μl [5,6]. These findings and the prognostic score proposed in this study emphasize the need for a specialized treatment strategy tailored to patients with HRL. We hope that, with the accumulation of more cases in the future, the optimal regimen for patients with HRL will be more clearly defined.

The current investigation has several limitations. First, this study is subject to potential selection bias, primarily owing to its reliance on personal experience and a limited number of case reports. This may affect the universal applicability of the findings. Additionally, although the proposed model marks a significant step towards creating a prognosis prediction tailored for the Japanese clinical landscape, it remains foundational and demands further refinement. Second, this study did not include specific lymphomas such as plasmablastic lymphoma, primary effusion lymphoma, and Hodgkin lymphoma. Given their prevalence among people with HIV in Japan [11–13], there is a clear need for specialized prediction models for these subtypes. Third, both the limited sample size and certain statistical measures in this study, notably high P values, might compromise the robustness of our conclusions and indicate that larger scale research could produce different results. Fourth, the development of the HIV-JPI leveraged the random forest algorithm to impute missing values, whereas the SPI was crafted employing listwise deletion. This resulted in slight variations between the elements of the HIV-JPI and SPI, stemming from the omission of cases with missing data in the construction of the latter. These nuances regarding missing data imputation in our study imply a potential avenue for additional validation. Future research should focus on these areas to enhance HRL prognosis predictions in Japan.

In conclusion, this analysis of Japanese patients with HRL highlighted patients’ age, HIV-RNA levels, EBV expression, and Ann Arbor classification as pivotal prognostic factors for HRL. The identification of HRL-specific prognostic markers offers the potential for stratified treatments, thereby elevating the clinical management. Although this investigation is grounded in a limited dataset, there lies a pressing imperative for expanded studies encompassing larger patient cohorts. Such endeavors hold promise for the development of an intricate and more precise clinical prediction model in the near future.

Acknowledgements

We express our profound gratitude to all the doctors, nurses, and other healthcare professionals at Yokohama Municipal Citizen's Hospital for providing dedicated patient care. We would like to express our profound gratitude to all the medical professionals across Japan who have tirelessly devoted themselves to the treatment and care of patients with HRL. We extend our profound gratitude to Masato Bingo and Tomoko Yamaguchi from Tokyo Medical University Hospital for referring a patient to us and for their valuable advice regarding this study.

Data sharing statement: the datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.

Conflicts of interest

There are no conflicts of interest.