Kaposi sarcoma (KS) is among the most common HIV-associated malignancies worldwide and is a leading cause of cancer in several African countries.1,2 KS is commonly staged using the AIDS Clinical Trials Group (ACTG) criteria, which classifies 3 variables—tumor extent (T), immune status (I), and systemic symptoms (S)—into good risk (0) and poor risk (1) groups. The ACTG staging criteria was originally proposed in 1989 and subsequently validated in 1997.3,4 Further studies conducted after the widespread availability of antiretroviral therapy (ART) suggested refinements to the original staging system, including lowering the CD4 T-cell count cutoff defining poor risk immune status.5,6
Although validated in the United States and Europe, the ACTG criteria have not been systematically evaluated in sub-Saharan Africa (SSA) or other resource-limited settings, where the burden of KS is greatest. Importantly, several features of HIV-associated KS disease presentation and management in SSA differ from Western countries, which could impact the prognostic value of the staging criteria. First, KS is typically aggressive in Africa, and patients often present late with widely disseminated and rapidly progressive disease.7 Second, the high rate of KS in African women contrasts sharply with the extremely low KS incidence among women in resource-rich countries.8–13 Third, patients frequently present with advanced HIV and other comorbid diseases, including tuberculosis, which may impact survival.14 Finally, KS outcomes in Africa are quite poor compared with the United States and Europe, with 5-year survival rates less than 10% in some regions.15 These differential outcomes may be because of both biologic factors as well as differences in resources available to manage the disease, including timely access to ART. Because of the unique features of KS presentation and management in SSA, we sought to determine whether the ACTG staging criteria accurately predicted survival among Ugandan patients with HIV-associated KS and whether we could construct a refined KS staging model for SSA.
Study Participants and Setting
We evaluated adults with HIV-associated KS who received care at the Uganda Cancer Institute (UCI) in Kampala, Uganda, the country's sole national cancer referral hospital. Patients were eligible for the study if they presented for care between September 1, 1992, and July 31, 2007, were ≥18 years of age, had histologically or clinically diagnosed KS, and had documentation of HIV-1 infection. Only patients who had at least one clinic visit in the 6-month period after KS diagnosis and sufficient data in their medical record to determine components of ACTG staging, including at least one documented CD4 count and a description of KS tumor extent at presentation, were included in the analysis.
Demographic and clinical data were obtained by review of the medical chart using a standardized case report form. Vital status at 2 years was abstracted from the medical chart when available; otherwise, we attempted to directly contact the patient or next of kin by calling the telephone contacts listed in the medical record to determine the patient's vital status up to 2 years from KS diagnosis.
KS presentation was based on variables described at the initial clinic visit at the UCI. Demographic variables included patient age and gender, and baseline clinical characteristics included KS lesion morphotype, presence of edema, and disease comorbidities. KS stage was defined according to the ACTG classification.3 Tumor stage was defined as T1 (poor risk) if there was tumor-associated edema or ulceration, oral involvement beyond the hard palate, or visceral involvement; otherwise stage was classified as T0 (good risk). Immune status was defined as poor risk (I1) if CD4 T-cell count was <200 cells per milliliter and good risk (I0) if ≥200 cells per milliliter. Systemic symptoms were defined as poor risk (S1) if the patient had a history of any opportunistic infections, presence of “B” symptoms, such as ≥10% weight loss, unexplained fevers and night sweats, or Karnofsky performance status <70. Treatments received at presentation or after KS diagnosis, including receipt of ART, chemotherapy, or radiation therapy, were also abstracted from the medical record.
We estimated survival using Kaplan–Meier methodology. Survival time was defined as the number of days from the date of clinical or histological diagnosis of KS (time 0) to the date of death, date of last documented clinic visit, or 2 years after KS diagnosis, whichever came first. Observations were censored at 2 years because of the paucity of clinical information beyond that time and to limit the potential effect of patients with longer follow-up on the overall risk estimates.
Survival curves for the different ACTG staging classifications were compared using log-rank tests. For evaluation of the immune status staging criteria, only baseline CD4 counts obtained within 180 days before or at KS diagnosis were used to determine immune status to avoid immortal time bias that can arise when variables measured after time 0 are used to determine baseline covariate status.16,17 Because baseline CD4 counts were only available for a subset of the cohort, we compared participants with and without baseline CD4 counts using chi-square or Fisher exact tests for categorical baseline factors and the Mann–Whitney test for continuous baseline factors to determine if any differences in these groups might bias our findings. To identify CD4 count classifications most predictive of survival, we evaluated the ACTG immune status CD4 cutoff (CD4 <200 vs CD4 ≥200), and 2 other parameterizations: (1) CD4 < 100 vs CD4 ≥100, shown to have better predictive value in some studies,5 and (2) continuous CD4. We used Cox proportional hazards (PH) models to estimate unadjusted and adjusted associations of CD4 count with survival among this subset of participants with baseline CD4 data. Model estimates were presented as hazard ratios (HR) with 95% confidence intervals (CIs).
Next, we constructed a multivariate model for predicting survival using baseline demographic and clinical variables, including age, sex, ACTG stage parameters (T, I, and S), presence of edema, and receipt of ART. We randomly selected two-thirds of the full cohort (training sample) to develop the multivariate Cox model and reserved the other one-third of the cohort (validation sample) to test the predictive ability of the model. This method should avoid over-optimism associated with selecting and evaluating the model using the same data. The functional form for age was explored using Martingale residuals. Although the residuals were not linear across the entire age range, a linear pattern within those aged 18–44 years and a separate linear pattern within those aged 45–74 years seemed plausible. We therefore used linear splines with a knot at age 45 years to allow for separate estimates in the 2 age groups. To assess the PH assumption, we tested for interactions between each baseline covariate and the logarithm of analysis time and used methods based on cumulative sums of martingale residuals.18 Because we found that many of the variables violated PH, we computed separate HR estimates for an early period (0–4 months post-KS diagnosis) during which chemotherapy is usually received and a late period (4–24 months post-KS diagnosis) after which initial chemotherapy is usually received. We postulated that the effect of our baseline covariates may vary depending on whether a patient is first receiving care or has already survived through initial chemotherapy.
Using the training sample, backward selection was used to obtain a final multivariate model with variables significant at the 0.05 level in at least one of the time periods, and baseline survival estimates were obtained at 4 months, 1 year, and 2 years post-KS diagnosis. These estimates were then used to compute model-predicted survival probabilities for each individual in the validation data set, at each of the 3 time points. To assess the model's predictive ability, we evaluated the receiver operating characteristics of the model-predicted survival probabilities at the 3 time points and computed the area under the curve (AUC).19 Two-sided P ≤ 0.05 were considered statistically significant. All statistical calculations were performed using SAS 9.3 and R version 3.1.1 software.
Approval for the study was obtained from the Makerere University School of Medicine Ethics and Research Committee; Fred Hutchinson Cancer Research Center IRB also approved participation of authors based at this institution (W.P., E.M.K., and C.C.). A waiver to contact patients, their next of kin, and referral HIV clinics was obtained to determine vital status for those participants lost to follow-up.
A total of 404 patients with HIV-associated KS met the eligibility criteria for the study. The median age was 35 years (range, 18–74 years) and more than half were men (53%) (Table 1). Among those with CD4 T-cell count documented within 180 days before KS diagnosis (N = 155), the median CD4 T-cell count was 69 cells per microliter (interquartile range, 14–194 cells/μL). Patients presented with a range of KS lesion morphotypes. Clinicians did not routinely note the presence of macular lesions, as these are the most common presentation of KS in Uganda; however, 213 participants (53%) were noted to have nodular lesions, 41 (10%) to have fungating lesions, and 108 (27%) to have both nodular and fungating lesions. KS-associated edema was observed among 235 patients (58%). By the ACTG staging classification, 321 patients (80%) had T1 tumor stage at diagnosis and 273 (68%) had systemic symptoms meeting poor risk criteria (S1). Among those with CD4 T-cell count data available at diagnosis, 118 patients (76%) had CD4 T-cell count below 200 cells per microliter and were classified as poor risk immune status (I1).
During the observation period, 306 patients (76%) received chemotherapy, which was almost exclusively combination chemotherapy with bleomycin and vincristine. At KS diagnosis, 165 patients (41%) were receiving ART, which was generally fixed-dose combination stavudine, lamivudine, and nevirapine (Triomune) or a lamivudine/zidovudine (Combivir)-based 3-drug regimen. An additional 59 patients (15%) received ART after KS diagnosis, during the observation period. Only 23 patients (6%) received radiation therapy.
At 2 years after KS diagnosis, 167 patients were alive, 156 had died, and 81 were lost to follow-up. The Kaplan–Meier estimate of 1-year overall survival was 65% (95% CI: 60% to 70%) and 2-year survival was 57% (95% CI: 51% to 62%). Among those who died, 44% died within 4 months of KS diagnosis. For patients with unknown vital status, 28% were lost to follow-up by 4 months. The median overall survival of the cohort was not reached by 2 years (Fig. 1). Survival curves differed significantly by tumor stage (P < 0.001) and by systemic symptoms (P < 0.001), but not by immune status (P = 0.57). By ACTG classification, the estimated 2-year survival percentage was 82% for good risk tumor (T0) compared with 50% for poor risk tumor (T1); 51% for good risk immune status (I0) compared with 49% for poor risk immune status (I1); and 69% for good risk systemic symptoms (S0) compared with 51% for poor risk systemic symptoms (S1) (Fig. 1; Table 2).
Association of CD4 Count With Survival
To identify CD4 count classifications most predictive of survival, we restricted the analysis to patients with baseline CD4 counts obtained within 180 days before KS diagnosis (n = 155). To assess potential selection bias, we compared characteristics of patients with and without baseline CD4 T-cell count data. The groups were similar with respect to age (P = 0.10), gender (P = 0.43), T1 tumor stage (P = 0.91), systemic symptoms (P = 0.30), edema (P = 0.09), lesion morphotype (P = 0.26), and receipt of radiation (P = 0.20). Those with baseline CD4 T-cell count data had a greater proportion receiving ART at KS diagnosis (P = 0.008) and a lower proportion receiving chemotherapy (P = 0.003) compared with those without baseline CD4 data. In addition to the ACTG immune status classification, we also explored 2 alternate parameterizations of baseline CD4 count for their associations with survival. Because of violations of PH, factors associated with survival were assessed for 2 time periods: 0–4 months and 4–24 months after KS diagnosis (Table 3). In unadjusted analysis, ACTG immune status (CD4 T-cell count <200 vs ≥200) was not significantly associated with survival in the first 4 months after KS diagnosis (HR = 1.8, 95% CI: 0.7 to 4.2, P = 0.21) or in months 4–24 after KS diagnosis (HR = 0.8, 95% CI: 0.4 to 1.7, P = 0.62). Similarly, CD4 count <100 vs ≥100 and continuous CD4 count were not significant in unadjusted analysis, for either time periods. We observed similar results for all 3 CD4 variables after adjusting for age, gender, ACTG tumor stage and systemic symptoms, and ART status at KS diagnosis (Table 3).
Predictive Model for Survival
To develop a predictive survival model, we randomly selected two-thirds of the full cohort as the training sample to construct Cox models with separate estimates for 2 time periods: 0–4 months and 4–24 months after diagnosis (Table 4). Because CD4 T-cell count was not significantly associated with survival in either unadjusted or adjusted analyses, we excluded this variable from further predictive models to have the full cohort available for analysis.
In unadjusted models, men had lower risk of death in the 4 months after diagnosis (HR = 0.5, 95% CI: 0.3 to 1.0, P = 0.04), but an increased risk of death in months 4–24 (HR = 1.8, 95% CI: 1.0 to 3.1, P = 0.046). The relationship between age and survival was not linear across the entire age range; so, separate estimates were presented for those aged 18–44 years and 45–74 years. Among those aged 18–44 years, older age was associated with lower risk of death in months 0–4 (HR per 10 years older = 0.5, 95% CI: 0.3 to 0.8, P = 0.002) but not in months 4–24. Among those aged 45–74 years, age was not significantly associated with survival in unadjusted analysis. Poor risk ACTG tumor stage (T1) was associated with greater risk of death in the 4- to 24-month period only (HR = 4.9, 95% CI: 1.8 to 13.5, P = 0.002). Poor risk systemic symptoms (S1) was strongly associated with greater risk of death in months 0–4 (HR = 7.4, 95% CI: 2.3 to 23.9, P < 0.001) and in months 4–24, albeit with a HR of lower magnitude (HR = 1.9, 95% CI: 1.1 to 3.5, P = 0.03). Presence of edema and ART status at the time of KS diagnosis were not significantly associated with survival in either time period.
The final multivariate model obtained using the training data set included age, ACTG tumor stage, and systemic symptoms. Gender did not remain significant in multivariate analyses and thus was not retained in the final model (0–4 months: HR for men versus women, adjusting for age, tumor stage, and systemic symptoms = 0.6, 95% CI: 0.3 to 1.1, P = 0.09; 4–24 months: HR = 1.5, 95% CI: 0.9 to 2.7, P = 0.15). Adjusted age estimates were similar to the unadjusted estimates. Poor risk ACTG tumor was not linked to survival in months 0–4, but was associated with greater risk of death in months 4–24 (HR = 4.0, 95% CI: 1.4 to 11.5, P = 0.009). Conversely, poor risk systemic symptoms was significantly associated with greater risk of death in months 0–4 (HR = 6.4, 95% CI: 1.9 to 21.1, P = 0.003), but less important in months 4–24 (HR = 1.6, 95% CI: 0.8 to 2.9, P = 0.16).
We assessed the predictive ability of the final model, using the validation data set (one-third of full cohort). The model best predicted survival status at 4 months after KS diagnosis (AUC = 0.598), compared with 1-year (AUC = 0.526) and 2-year post-KS diagnosis (AUC = 0.508) (Figure 1, Supplemental Digital Content, http://links.lww.com/QAI/A965). However, the model's predictive ability was limited at all 3 time points, evidenced by the relatively low AUC values.
This is the first study to our knowledge to systematically evaluate the ACTG KS staging criteria among patients in SSA, where KS burden is greatest. Overall survival was poor in the cohort, with a 2-year survival estimate of only 57%. We found that the risk classifications for Tumor (T) and Systemic symptoms (S) had prognostic value, but their ability to predict survival differed based on the time period following KS diagnosis. The presence of systemic symptoms was associated with “early” death in the first 4 months of treatment, whereas high tumor burden was associated with “late” death between 4 months and 2 years after diagnosis. This difference in predictive value between early and late periods is intriguing and suggests that different biologic processes may influence outcomes in these periods following KS diagnosis. Nearly half of all KS patient deaths occurred in the first 4 months of treatment, and it is plausible that factors reflected in the poor risk systemic symptom criteria, which include history of opportunistic infection, “B” symptoms, and poor performance status, contributed to death in the early phase of treatment. These poor risk factors could represent advanced KS, comorbid disease, including tuberculosis, cryptococcosis, and other common opportunistic infections in SSA, or a proinflammatory state in the setting of KS, such as KS-IRIS or the KSHV Inflammatory Cytokine Syndrome.20 Any of these conditions could limit a patient's ability to respond to chemotherapy; however, if a patient does survive the initial treatment, baseline tumor burden may be a more important determinant of longer term survival. Additional studies are needed to test these hypotheses, but this observation may indicate the need for tailored treatment strategies for KS patients at high risk for early death.
Baseline CD4 count was not a predictor of KS survival, which has also been observed in other studies that have evaluated predictors of KS survival since the widespread use of ART.5,21 This finding may indicate that with increasing availability of ART, CD4 at diagnosis is less important for overall survival than it was when the ACTG criteria were originally described in the pre-ART era. Interestingly, we did not find that receipt of ART was associated with survival in our cohort, which has been reported by others.5,22 Treatment selection bias may have attenuated this observation if patients with the lowest CD4 counts were most likely to receive ART during the period of observation. In addition, because ART is generally managed by HIV clinics in Uganda, cancer center records may not have adequately captured ART administration at outside clinics and underestimated ART receipt in the cohort; we also did not have access to adherence records. Furthermore, ART regimens used in Uganda during the study period were not as potent as other regimens currently available and may not have achieved HIV viral suppression in all patients. Finally, our final multivariate model also identified age as a predictor of survival, with younger age associated with death in the first 4 months of treatment and older age associated with death after 4 months. The reason for the differential association with age is unclear, but it could reflect more aggressive disease in the youngest patients and age-related comorbidities in older patients.
Importantly, our new model, which included T, S, and age, was not very good at predicting survival, which suggests that other factors not measured in the model may be important determinants of outcomes. A few studies have attempted to identify additional predictors of KS treatment response, including CD8+ T-cell counts and HIV viral load levels, but no factor has been consistently associated with KS improvement or survival.5,23–25 High pretreatment plasma human herpesvirus 8 quantity has also been associated with poor outcomes,6,21 but the appropriate parameterization and predictive value of plasma human herpesvirus 8 needs to be evaluated in other cohorts. Current prospective KS studies in Uganda and elsewhere in SSA will help to validate these observations and identify additional prognostic and treatment selective biomarkers.
As a retrospective chart review, our study was limited by loss to follow-up and incomplete data in the medical records. In particular, many KS patients treated at the UCI were not eligible for the study because the data in the medical chart were not sufficient to assign ACTG staging. In addition, baseline CD4 within the first 180 days before KS diagnosis was only available in a subset of participants, which may have limited our ability to detect an association between CD4 count and survival. Similarly, other clinical parameters, including HIV viral load or laboratory measures of organ function, were not uniformly available in the medical charts and could not be assessed. We also did not evaluate socioeconomic factors, such as income or education level, on survival outcomes. Finally, treatment provided, particularly ART, was not consistent across the cohort, in part because of early scale-up of ART access in Uganda during the study period. This variability in receipt of ART may have limited our ability to determine its effect on survival, but it does reflect current gaps in treatment access that exist for many KS patients in SSA and consequently provides a realistic context for evaluating staging criteria in Africa.
Despite these limitations, our study demonstrates that ACTG criteria for tumor burden and systemic symptoms are associated with survival among adults with HIV-associated KS in SSA; however, the limited predictive value of our multivariate model demonstrates the need to identify additional biomarkers to more reliably stage KS in Africa. Furthermore, the poor 1-year and 2-year overall survival in our cohort demands better implementation of existing treatment strategies, including improved access to timely ART and chemotherapy, and the development of novel treatment interventions to improve outcomes among patients with KS in SSA.
The authors thank the study participants, their families, and the study team, whose commitment made this study possible.
1. Ferlay J, Soerjomataram I, Dikshit R, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer. 2015;136:E359–E386.
2. Wabinga HR, Nambooze S, Amulen PM, et al. Trends in the incidence of cancer in Kampala, Uganda 1991–2010. Int J Cancer. 2014;135:432–439.
3. Krown SE, Metroka C, Wernz JC. Kaposi's sarcoma in the acquired immune deficiency syndrome: a proposal for uniform evaluation, response, and staging
criteria. AIDS Clinical Trials Group Oncology Committee. J Clin Oncol. 1989;7:1201–1207.
4. Krown S, Testa M, Huang J. AIDS-related Kaposi's sarcoma: prospective validation of the AIDS Clinical Trials group staging
classification. AIDS clinical Trials group Oncology Committee. J Clin Oncol. 1997;15:3085–3092.
5. Nasti G, Talamini R, Antinori A, et al. AIDS-related Kaposi's sarcoma: evaluation of potential new prognostic factors and assessment of the AIDS clinical trial group staging
system in the Haart Era—the Italian Cooperative Group on AIDS and Tumors and the Italian Cohort of Patients Naive From Antiretrovirals. J Clin Oncol. 2003;21:2876–2882.
6. El Amari E, Toutous-Trellu L, Gayet-Ageron A. Predicting the evolution of Kaposi sarcoma
in the highly active antiretroviral therapy era. AIDS. 2008;22:1019–1028.
7. Mosam A, Shaik F, Uldrick TS, et al. A randomized controlled trial of highly active antiretroviral therapy versus highly active antiretroviral therapy and chemotherapy in therapy-naive patients with HIV-associated Kaposi sarcoma
in South Africa. J Acquir Immune Defic Syndr. 2012;60:150–157.
8. Lassoued K, Clauvel JP, Fegueux S, et al. AIDS-associated Kaposi's sarcoma in female patients. AIDS. 1991;5:877–880.
9. Cooley TP, Hirschhorn LR, O'Keane JC. Kaposi's sarcoma in women with AIDS. AIDS. 1996;10:1221–1225.
10. Nasti G, Serraino D, Ridolfo A, et al. AIDS-associated Kaposi's sarcoma is more aggressive in women: a study of 54 patients. J Acquir Immune Defic Syndr Hum Retrovirol. 1999;20:337–341.
11. Meditz AL, Borok M, MaWhinney S, et al. Gender differences in AIDS-associated Kaposi sarcoma
in Harare, Zimbabwe. J Acquir Immune Defic Syndr. 2007;44:306–308.
12. Mosam A, Hurkchand HP, Cassol E, et al. Characteristics of HIV-1-associated Kaposi's sarcoma among women and men in South Africa. Int J STD AIDS. 2008;19:400–405.
13. Phipps W, Ssewankambo F, Nguyen H, et al. Gender differences in clinical presentation and outcomes of epidemic Kaposi sarcoma
in Uganda. PLoS One. 2010;5:e13936.
14. Agaba PA, Sule HM, Ojoh RO, et al. Presentation and survival of patients with AIDS- related Kaposi's sarcoma in Jos, Nigeria. Int J STD AIDS. 2009;20:410–413.
15. Gondos A, Brenner H, Wabinga H, et al. Cancer survival in Kampala, Uganda. Br J Cancer. 2005;92:1808–1812.
16. Wolfe RA, Strawderman RL. Logical and statistical fallacies in the use of Cox regression models. Am J Kidney Dis. 1996;27:124–129.
17. Suissa S. Immortal time bias in observational studies of drug effects. Pharmacoepidemiol Drug Saf. 2007;16:241–249.
18. Lin DY, Wei LJ, Ying Z. Checking the Cox model with cumulative sums of martingale-based residuals. Biometrika. 1993;80:557–572.
19. Heagerty PJ, Lumley T, Pepe MS. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics. 2000;56:337–344.
20. Uldrick TS, Wang V, O'Mahony D, et al. An interleukin-6-related systemic inflammatory syndrome in patients co-infected with Kaposi sarcoma
-associated herpesvirus and HIV but without Multicentric Castleman disease. Clin Infect Dis. 2010;51:350–358.
21. Borok M, Fiorillo S, Gudza I, et al. Evaluation of plasma human herpesvirus 8 DNA as a marker of clinical outcomes during antiretroviral therapy for AIDS-related Kaposi sarcoma
in Zimbabwe. Clin Infect Dis. 2010;51:342–349.
22. Stebbing J, Sanitt A, Nelson M, et al. A prognostic index for AIDS-associated Kaposi's sarcoma in the era of highly active antiretroviral therapy. Lancet. 2006;367:1495–1502.
23. Nasti G, Martellotta F, Baretta M, et al. Impact of highly active antiretroviral therapy on the presenting features and outcome of patients with acquired immunodeficiency syndrome-related Kaposi. Cancer. 2003;98:2440–2446.
24. Tam HK, Zhang ZF, Jacobson LP, et al. Effect of highly active antiretroviral therapy on survival among HIV-infected men with Kaposi sarcoma
or non-Hodgkin lymphoma. Int J Cancer. 2002;98:916–922.
25. Stebbing J, Sanitt A, Teague A, et al. Prognostic significance of immune subset measurement in individuals with AIDS-associated Kaposi's sarcoma. J Clin Oncol. 2007;25:2230–2235.
Kaposi sarcoma; staging; sub-Saharan Africa
Supplemental Digital Content
Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.