Showing papers by "Moses R. Kamya published in 2012"

Neurocognitive and Motor Deficits in HIV-Infected Ugandan Children With High CD4 Cell Counts

Abstract: Human immunodeficiency virus (HIV) has been shown to cause both cognitive and motor dysfunction in perinatally infected children [1–4]. Severe compromise of the immune system from HIV is associated with high rates of neurodevelopmental disability [5], but little is known about the function of HIV-infected children before they have experienced a significant decrease in CD4 lymphocyte cell counts. Studies of neurodevelopmental function in HIV-infected African children have been small and have generated conflicting results [6–9]. One report of 26 Congolese HIV-infected school-aged children demonstrated cognitive, language, and motor impairment using the first-edition Kaufman Assessment Battery for Children (KABC), compared with HIV-uninfected exposed and control children [10]. In contrast, Bagenda et al [11] reported no significant impairment among 28 HIV-infected ART-naive Ugandan school children, except in one KABC subscale of hand movements and in the Wide Range Achievement Test - Third Edition of academic achievement. Neither study was able to investigate the degree of impairment in relation to quantitative measures of disease stage such as HIV RNA level or CD4 cell count. As access to HIV testing and care expands throughout Africa, increasing numbers of school-age HIV-infected children are being identified. Many of these children are ineligible for antiretroviral therapy (ART) because of high CD4 cell counts. It is important to know whether these children suffer impairment of their neurocognitive or motor development. In this study, we sought to determine whether HIV-infected ART-naive Ugandan children with CD4 cell measures that were above World Health Organization (WHO) [12] thresholds for ART initiation exhibited neurocognitive or motor deficits compared with HIV-uninfected controls from 2 observational cohorts in Kampala, Uganda [12]. We also examined the associations of WHO clinical stage, CD4 cell count and percentage, and plasma HIV RNA level with neurodevelopmental deficits.

Antiretroviral Agents and Prevention of Malaria in HIV-Infected Ugandan Children

Abstract: A B S T R AC T Background Human immunodeficiency virus (HIV) protease inhibitors show activity against Plasmodium falciparum in vitro. We hypothesized that the incidence of malaria in HIVinfected children would be lower among children receiving lopinavir–ritonavir– based antiretroviral therapy (ART) than among those receiving nonnucleoside reverse-transcriptase inhibitor (NNRTI)–based ART. Methods We conducted an open-label trial in which HIV-infected children 2 months to 5 years of age who were eligible for ART or were currently receiving NNRTI-based ART were randomly assigned to either lopinavir–ritonavir–based ART or NNRTI-based ART and were followed for 6 months to 2 years. Cases of uncomplicated malaria were treated with artemether–lumefantrine. The primary end point was the incidence of malaria. Results We enrolled 176 children, of whom 170 received the study regimen: 86 received NNRTI-based ART, and 84 lopinavir–ritonavir–based ART. The incidence of ma laria was lower among children receiving the lopinavir–ritonavir–based regimen than among those receiving the NNRTI-based regimen (1.32 vs. 2.25 episodes per person-year; incidence-rate ratio, 0.59; 95% confidence interval [CI], 0.36 to 0.97; P = 0.04), as was the risk of a recurrence of malaria after treatment with artemether–lumefantrine (28.1% vs. 54.2%; hazard ratio, 0.41; 95% CI, 0.22 to 0.76; P = 0.004). The median lumefantrine level on day 7 after treatment for malaria was significantly higher in the lopinavir–ritonavir group than in the NNRTI group. In the lopinavir–ritonavir group, lumefantrine levels exceeding 300 ng per milliliter on day 7 were associated with a reduction of more than 85% in the 63-day risk of recurrent malaria. A greater number of serious adverse events occurred in the lopinavir– ritonavir group than in the NNRTI group (5.6% vs. 2.3%, P = 0.16). Pruritus occurred significantly more frequently in the lopinavir–ritonavir group, and elevated alanine aminotransferase levels significantly more frequently in the NNRTI group. Conclusions Lopinavir–ritonavir–based ART as compared with NNRTI-based ART reduced the incidence of malaria by 41%, with the lower incidence attributable largely to a significant reduction in the recurrence of malaria after treatment with artemether– lumefantrine. Lopinavir–ritonavir–based ART was accompanied by an increase in serious adverse events. (Funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development; ClinicalTrials.gov number, NCT00978068.)

Indoor Residual Spraying of Insecticide and Malaria Morbidity in a High Transmission Intensity Area of Uganda

Abstract: Background Recently the use of indoor residual spraying of insecticide (IRS) has greatly increased in Africa; however, limited data exist on the quantitative impacts of IRS on health outcomes in highly malaria endemic areas.

Increasing incidence of malaria in children despite insecticide-treated bed nets and prompt anti-malarial therapy in Tororo, Uganda

Abstract: Background The burden of malaria has decreased in parts of Africa following the scaling up of control interventions. However, similar data are limited from high transmission settings.

The association between malnutrition and the incidence of malaria among young HIV-infected and -uninfected Ugandan children: a prospective study

Abstract: Background: In sub-Saharan Africa, malnutrition and malaria remain major causes of morbidity and mortality in young children. There are conflicting data as to whether malnutrition is associated with an increased or decreased risk of malaria. In addition, data are limited on the potential interaction between HIV infection and the association between malnutrition and the risk of malaria. Methods: A cohort of 100 HIV-unexposed, 203 HIV-exposed (HIV negative children born to HIV-infected mothers) and 48 HIV-infected children aged 6 weeks to 1 year were recruited from an area of high malaria transmission intensity in rural Uganda and followed until the age of 2.5 years. All children were provided with insecticidetreated bed nets at enrolment and daily trimethoprim-sulphamethoxazole prophylaxis (TS) was prescribed for HIVexposed breastfeeding and HIV-infected children. Monthly routine assessments, including measurement of height and weight, were conducted at the study clinic. Nutritional outcomes including stunting (low height-for-age) and underweight (low weight-for-age), classified as mild (mean z-scores between -1 and -2 during follow-up) and moderate-severe (mean z-scores < -2 during follow-up) were considered. Malaria was diagnosed when a child presented with fever and a positive blood smear. The incidence of malaria was compared using negative binomial regression controlling for potential confounders with measures of association expressed as an incidence rate ratio (IRR). Results: The overall incidence of malaria was 3.64 cases per person year. Mild stunting (IRR = 1.24, 95% CI 1.061.46, p = 0.008) and moderate-severe stunting (IRR = 1.24, 95% CI 1.03-1.48, p = 0.02) were associated with a similarly increased incidence of malaria compared to non-stunted children. Being mildly underweight (IRR = 1.09, 95% CI 0.95-1.25, p = 0.24) and moderate-severe underweight (IRR = 1.12, 95% CI 0.86-1.46, p = 0.39) were not associated with a significant difference in the incidence of malaria compared to children who were not underweight. There were no significant interactions between HIV-infected, HIV-exposed children taking TS and the associations between malnutrition and the incidence of malaria. Conclusions: Stunting, indicative of chronic malnutrition, was associated with an increased incidence of malaria among a cohort of HIV-infected and -uninfected young children living in an area of high malaria transmission intensity. However, caution should be made when making causal inferences given the observational study design and inability to disentangle the temporal relationship between malnutrition and the incidence of malaria. Trial Registration: ClinicalTrials.gov: NCT00527800.

Electronic Medical Records and Same Day Patient Tracing Improves Clinic Efficiency and Adherence to Appointments in a Community Based HIV/AIDS Care Program, in Uganda

Abstract: Patients who miss clinic appointments make unscheduled visits which compromise the ability to plan for and deliver quality care. We implemented Electronic Medical Records (EMR) and same day patient tracing to minimize missed appointments in a community-based HIV clinic in Kampala. Missed, early, on-schedule appointments and waiting times were evaluated before (pre-EMR) and 6 months after implementation of EMR and patient tracing (post-EMR). Reasons for missed appointments were documented pre and post-EMR. The mean daily number of missed appointments significantly reduced from 21 pre-EMR to 8 post-EMR. The main reason for missed appointments was forgetting (37%) but reduced significantly by 30% post-EMR. Loss to follow-up (LTFU) also significantly decreased from 10.9 to 4.8% The total median waiting time to see providers significantly decreased from 291 to 94 min. Our findings suggest that EMR and same day patient tracing can significantly reduce missed appointments, and LTFU and improve clinic efficiency.

Maternal nutritional status predicts adverse birth outcomes among HIV-infected rural Ugandan women receiving combination antiretroviral therapy.

Abstract: Objective: Maternal nutritional status is an important predictor of birth outcomes, yet little is known about the nutritional status of HIV-infected pregnant women treated with combination antiretroviral therapy (cART). We therefore examined the relationship between maternal BMI at study enrollment, gestational weight gain (GWG), and hemoglobin concentration (Hb) among 166 women initiating cART in rural Uganda. Design: Prospective cohort. Methods: HIV-infected, ART-naive pregnant women were enrolled between 12 and 28 weeks gestation and treated with a protease inhibitor or non-nucleoside reverse transcriptase inhibitor-based combination regimen. Nutritional status was assessed monthly. Neonatal anthropometry was examined at birth. Outcomes were evaluated using multivariate analysis. Results: Mean GWG was 0.17 kg/week, 14.6% of women experienced weight loss during pregnancy, and 44.9% were anemic. Adverse fetal outcomes included low birth weight (LBW) (19.6%), preterm delivery (17.7%), fetal death (3.9%), stunting (21.1%), small-for-gestational age (15.1%), and head-sparing growth restriction (26%). No infants were HIV-infected. Gaining ,0.1 kg/week was associated with LBW, preterm delivery, and a composite adverse obstetric/fetal outcome. Maternal weight at 7 months gestation predicted LBW. For each g/dL higher mean Hb, the odds of small-for-gestational age decreased by 52%. Conclusions: In our cohort of HIV-infected women initiating cART during pregnancy, grossly inadequate GWG was common. Infants whose mothers gained ,0.1 kg/week were at increased risk for LBW, preterm delivery, and composite adverse birth outcomes. cART by itself may not be sufficient for decreasing the burden of adverse birth outcomes among HIV-infected women. Trial Registration: Clinicaltrials.gov NCT00993031

Factors associated with malaria parasitaemia, malnutrition, and anaemia among HIV-exposed and unexposed Ugandan infants: a cross-sectional survey

Abstract: Background Malaria, malnutrition and anaemia are major causes of morbidity and mortality in African children. The interplay between these conditions is complex and limited data exist on factors associated with these conditions among infants born to HIV-uninfected and infected women.

Prevalence and factors associated with cryptococcal antigenemia among severely immunosuppressed HIV-infected adults in Uganda: a cross-sectional study

Abstract: Background Cryptococcal infection is a common opportunistic infection among severely immunosuppressed HIV patients and is associated with high mortality. Positive cryptococcal antigenemia is an independent predictor of cryptococcal meningitis and death in patients with severe immunosuppression. We evaluated the prevalence and factors associated with cryptococcal antigenemia among patients with CD4 counts of 100 cells/mm3 or less in Mulago Hospital, Kampala, Uganda. Screening of a targeted group of HIV patients may enable early detection of cryptococcal infection and intervention before initiating antiretroviral therapy. Factors associated with cryptococcal antigenemia may be used subsequently in resource-limited settings in screening for cryptococcal infection, and this data may also inform policy for HIV care.

Potential for false positive HIV test results with the serial rapid HIV testing algorithm

Abstract: Background: Rapid HIV tests provide same-day results and are widely used in HIV testing programs in areas with limited personnel and laboratory infrastructure. The Uganda Ministry of Health currently recommends the serial rapid testing algorithm with Determine, STAT-PAK, and Uni-Gold for diagnosis of HIV infection. Using this algorithm, individuals who test positive on Determine, negative to STAT-PAK and positive to Uni-Gold are reported as HIV positive. We conducted further testing on this subgroup of samples using qualitative DNA PCR to assess the potential for false positive tests in this situation. Results: Of the 3388 individuals who were tested, 984 were HIV positive on two consecutive tests, and 29 were considered positive by a tiebreaker (positive on Determine, negative on STAT-PAK, and positive on Uni-Gold). However, when the 29 samples were further tested using qualitative DNA PCR, 14 (48.2%) were HIV negative. Conclusion: Although this study was not primarily designed to assess the validity of rapid HIV tests and thus only a subset of the samples were retested, the findings show a potential for false positive HIV results in the subset of individuals who test positive when a tiebreaker test is used in serial testing. These findings highlight a need for confirmatory testing for this category of individuals.

Virologic versus immunologic monitoring and the rate of accumulated genotypic resistance to first-line antiretroviral drugs in Uganda

Abstract: Background Viral load monitoring (VLM) to identify individuals failing antiretroviral therapy (ART) is not widely available in resource-limited settings. We compared the genotypic resistance patterns between clients with VLM versus immunological monitoring (IM).

Health facility-based malaria surveillance: The effects of age, area of residence and diagnostics on test positivity rates

Abstract: Background: The malaria test positivity rate (TPR) is increasingly used as an indicator of malaria morbidity because TPR is based on laboratory-confirmed cases and is simple to incorporate into existing surveillance systems. However, temporal trends in TPR may reflect changes in factors associated with malaria rather than true changes in malaria morbidity. This study examines the effects of age, area of residence and diagnostic test on TPR at two health facilities in regions of Uganda with differing malaria endemicity. Methods: The analysis included data from diagnostic blood smears performed at health facilities in Walukuba and Aduku between January 2009 and December 2010. The associations between age and time and between age and TPR were evaluated independently to determine the potential for age to confound temporal trends in TPR. Subsequently, differences between observed TPR and TPR adjusted for age were compared to determine if confounding was present. A similar analysis was performed for area of residence. Temporal trends in observed TPR were compared to trends in TPR expected using rapid diagnostic tests, which were modelled based upon sensitivity and specificity in prior studies. Results: Age was independently associated with both TPR and time at both sites. At Aduku, age-adjusted TPR increased relative to observed TPR due to the association between younger age and TPR and the gradual increase in age distribution. At Walukuba, there were no clear differences between observed and age-adjusted TPR. Area of residence was independently associated with both TPR and time at both sites, though there were no clear differences in temporal trends in area of residence-adjusted TPR and observed TPR at either site. Expected TPR with pLDH- and HRP-2-based rapid diagnostic tests (RDTs) was higher than observed TPR at all time points at both sites. Conclusions: Adjusting for potential confounders such as age and area of residence can ensure that temporal trends in TPR due to confounding are not mistakenly ascribed to true changes in malaria morbidity. The potentially large effect of diagnostic test on TPR can be accounted for by calculating and adjusting for the sensitivity and specificity of the test used.

The effect of HIV on malaria in the context of the current standard of care for HIV-infected populations in Africa

Abstract: HIV infection affects the clinical pattern of malaria. There is emerging evidence to suggest that previously documented interactions may be modified by recently scaled-up HIV and malaria interventions. Prophylaxis with trimethoprim–sulfamethoxazole (TS) in combination with use of insecticide-treated nets can markedly decrease the incidence of malaria in HIV-infected pregnant and nonpregnant adults and children even in the setting of antifolate resistance-conferring mutations that are currently common in Africa. Nonetheless, additional interventions are needed to protect HIV-infected people that reside in high-malaria-transmission areas. Artemether–lumefantrine and dihydroartemisinin–piperaquine are highly efficacious and safe for the treatment of uncomplicated malaria in HIV-infected persons. Coadministration of antiretroviral and antimalarial drugs creates the potential for pharmacokinetic drug interactions that may increase (causing enhancement of malaria treatment efficacy and post-treatment prophylaxi...

Regional differences in predictive accuracy of WHO immunologic failure criteria.

Abstract: We compared the performance of the WHO immunologic criteria for treatment failure among Uganda and American patients. Antiretroviral treatment-naive patients with a CD4 T-cell count less than 200 cells/µl or AIDS at enrollment on a nonnucleoside reverse transcriptase inhibitors-based regimen for more than 1 year were selected. For all criteria, the positive predictive value was significantly higher in the American compared with the Ugandan patients. Population-specific guidelines should be developed using large African cohorts to identify more specific and sensitive criteria. In industrialized countries, antiretroviral treatment (ART) efficacy is monitored through routine measurement of CD4+ cell counts and plasma viral load. Guidelines from the WHO on ART for HIV infection in resource-limited settings recommend the use of immunologic monitoring when viral load testing is not available [1,2]. Previous studies from sub-Saharan Africa have shown that the proposed criteria are neither sensitive nor specific [3–7]. We suspected that one of the reasons for poor performance in African patients is the high incidence of opportunistic infections, which could consequently impair the immune reconstitution of patients on ART [8,9]. We sought to evaluate the performance of the WHO immunologic criteria for treatment failure in a Ugandan and an American cohort and to compare the predictive values of these criteria in identifying patients on first-line ART with viral failure. We also examined the relative contribution of opportunistic infections on the performance of the criteria among the Ugandan patients. The Infectious Diseases Institute (IDI) Research Prospective Observational Cohort is a closed research cohort of 559 patients, enrolled at ART initiation between April 2004 and April 2005 in Uganda. Details of the cohort have been published elsewhere [10]. The Johns Hopkins HIV Clinical Cohort (JHHCC) is an open observational longitudinal cohort database of HIV-infected patients in the USA, established in 1989 with over 6000 patients enrolled, with majority initiated on ART between 1999 and 2003. Details of the cohort have been published elsewhere [11]. Patients selected from both cohorts were as follows: ART-naive at cohort enrollment; initiated on zidovudine/stavudine with lamivudine and efavirenz/nevirapine; baseline CD4+ cell count less than 200 cells/µl or clinical criteria of AIDS; and on ART for at least 1 year. We compared the characteristics of the patients at ART initiation in both cohorts. Categorical and continuous variables were compared using chi-square and Mann–Whitney tests, respectively. To compare the performance of the WHO criteria in both cohorts, we identified the proportion of patients fulfilling at least one of the three criteria: CD4+ cell count less than 100 cell/µl after 12 months; more than 50% drop from CD4+ cell count peak; and CD4+ cell count lower than baseline [1]. We then obtained the proportion of patients with confirmed viral failure at the time they met any of the criteria. We calculated the positive and negative predictive values (PPV and NPV). Finally, we did a sensitivity analysis excluding patients who had ever had an opportunistic infection in the first year of ART in the IDI cohort. Ethical approval was obtained from the local Institutional Review Board committees. We included 442 patients from the IDI cohort and 153 patients from the JHHCC. At ART initiation, a lower proportion of patients from the JHHCC were female (28.8 versus 69%, P < 0.0001), had a slightly lower median age [35 years, interquartile range (IQR) 30–41 years versus 39 years, IQR 35–44 years, P < 0.001], higher median BMI (22 kg/m2, IQR, 19–26 kg/m2 versus 20 kg/m2, IQR 17–22 kg/m2, P < 0.0001), lower median CD4+ cell count per microliter (42 versus 102, P = 0.064) and a higher proportion initiated on zidovudine with lamivudine and efavirenz (39% versus 26%, P = 0.012). A larger proportion of the IDI cohort experienced an opportunistic infection in the first year of ART (28.7% versus 13.7%, P = 0.0002). Tuberculosis was the most common opportunistic infection, with (4.7%) in the IDI cohort compared with (0.6%) in the JHHCC (P = 0.021). The proportion of patients who fulfilled each of the criteria and had confirmed virologic failure in the IDI cohort versus JHHCC were as follows: 1.1% versus 6.5% (P = 0.0002) for criterion 1, 2.9% versus 37.9% (P < 0.0001) for criterion 2 and 4.1% versus 5.9% (P = 0.354) for criterion 3. As shown in Table 1, the PPV for virologic failure in the IDI cohort and the JHHCC, respectively, was 13.5% and 52.6%, for criterion 1; 26.0% and 84.1%, for criteria 2; and 20.4% and 78.9%, for criteria 3. The NPV for criteria 1, 2 and 3 in the IDI cohort and JHHCC, respectively, was 87.6% versus 85.5%, 93.6% versus 84.4% and 93.5% versus 71.8%. Table 1 Proportion of patients fulfilling the immunological criteria for treatment failure and proportion of patients with confirmed virologic failure. In a sensitivity analysis including only IDI cohort patients (71.3%) who did not experience an opportunistic infection, there was no significant difference in the PPV of this subgroup compared with the entire cohort for criteria 1 and 2: criterion 1, 13.0% versus 13.5% (P = 0.835); criterion 2, 26.0% versus 20.4% (P = 0.06); but, a significant difference for criterion 3 (20.4% versus 10.2%, P < 0.0001). The WHO immunologic criteria selected higher proportions of virologically failing patients in the American than in a Ugandan cohort and had higher PPVs. Our results suggest that this difference is not due to the high rate of diagnosed opportunistic infections in the African setting, even though severe bacterial diseases often seen could also affect immunological response. Other possibilities include undiagnosed and untreated opportunistic infections, poor nutrition and differences in HIV subtypes. Studies from the east African region suggest that subtype D is associated with a faster decline in CD4+ cell count and increased risk of mortality compared with subtype A [12,13] due to increased CD4+ cells activation and subsequent apoptosis [14]. It is important to evaluate the impact of host characteristics of the infected population on ART efficacy and viral subtype [15]. Published data suggest that, with immunological monitoring only, patients with confirmed viral failure are identified after significant accumulation of drug resistance [16]. Recent studies have shown that a viral load will reduce an early and unnecessary switch to second-line therapy, reduce accumulation of resistance and prevent drug resistance viral transmission by patients who are failing [17–19]. Our study limitations included, first, the differences in the data used in both cohorts; however, patients were identified during the roll-out of ART in both countries. Second, the younger age of JHHCC patients compared with IDI cohort patients could have influenced the higher immune response in JHHCC, as has been shown previously [20,21]. In summary, our data highlight the need for viral load measurements when identifying treatment failure in HIV-positive individuals on ART. Further research into the risk factors for the limited performance of these WHO immunologic criteria in resource-limited settings is needed. Population-specific guidelines should be developed using large African cohorts to identify more specific and sensitive criteria.

Virologic suppression in nevirapine-exposed HIV-infected infants initiating antiretroviral therapy in rural Uganda.

Abstract: We measured virologic suppression among 34 nevirapine (NVP)-exposed HIV-infected children with median age of 8.6 months (range: 3.2–19.9) initiating NVP-based antiretroviral therapy (ART) in rural Uganda. In Kaplan–Meier analysis, the cumulative probability of virologic suppression, defined as having two consecutive HIV-1 RNA 7 50 000 copies ml−1, p = 0.03) and high CD4% (HR = 6.0 for CD4% > 25, p = 0.003). These results lend additional support to the 2010 World Health Organization recommendations that protease inhibitors be used to treat NVP-exposed children, but that NVP-based ART should be initiated before the decline of CD4% to optimize outcomes in NVP-exposed children when protease inhibitors are not available.

Response to comment on "prevalence and factors associated with cryptococcal antigenemia among severely immunosuppressed HIV-infected adults in Uganda (Oyella et al. 2012)".

Abstract: We would like to thank you for the comments regarding our recent article: Cryptococcal antigenemia among severely immunosuppressed HIV-infected adults In response to your comment: Sir, the recent report on ‘‘Prevalence and factors associated with cryptococcal antigenemia among severely immunosuppressed HIV-infected adults in Uganda’’ is very interesting Oyella et al concluded that ‘‘Independent predictors of positive serum cryptococcal antigenemia were CD4+ T cell counts of less than 50 cells/mm, low body mass index, neck pain, signs of meningeal irritation, and a recent diagnosis of HIV infection’’ This work is a cross sectional study, not a case-control study; hence, there might be some bias on assessment of risk factor (Published: 2 August 2012) Citation: Oyella J et al Journal of the International AIDS Society 2012, 15 :18003 http://wwwjiasocietyorg/indexphp/jias/article/view/18003 | http://dxdoiorg/107448/IAS15218003

Evaluation of the AIDS clinical trials group staging criteria for Kaposi Sarcoma in a resource limited setting

Abstract: Background Kaposi sarcoma (KS) is commonly staged using by the AIDS Clinical Trials Group (ACTG) criteria. The three variables of the ACTG are dichotomized as good risk (0) and poor risk (1). Good risk immune status (I0) is defined as CD4 T-cell count ≥200cells/µl, and poor risk (I1) as CD4 < 200cells/µl. Although validated in the US and Europe, no evaluation has been done in resourcelimited settings during the HAART era. We sought to determine whether the ACTG staging criteria is predictive of overall survival among Ugandan patients with HIV-associated KS. Methods Data were abstracted from medical records of adult patients with HIV-associated KS seen at the Uganda Cancer Institute (UCI) from 2000-2006. The primary outcome was 2-year overall survival. Vital status at 2 years was determined from the medical chart, or by contacting the patient or next of kin using the phone contact provided in the chart or ART clinic. Survival was modeled using Kaplan-Meier methods. Factors associated with survival were evaluated using Cox proportional hazards.