As a result of antiretroviral treatment (ART), patients with HIV now have a life expectancy that is approaching that of the general population,1 and deaths amongst those with HIV are now more likely to occur from non-AIDS events than AIDS itself.2,3 As a result, the focus of treatment for HIV-positive individuals has shifted from prolonging short-term survival to reducing long-term HIV-induced and drug-induced adverse events.
Several biomarkers are routinely measured in those living with HIV, either because abnormal values of these biomarkers provide an early indication of antiretroviral drug–related side effects, or because the biomarkers are associated with poor outcomes in the HIV-negative population more generally. These include liver enzymes [eg, alanine aminotransferase (ALT), alkaline phosphotase (ALK), bilirubin (BIL)] and markers of kidney function [creatinine (CRE) and urea (URE)]. Patients receiving some antiretroviral drugs have also been shown to be at increased risk of myocardial infarction.4 Other markers, such as low albumin (ALB) have been shown to be associated with mortality in people with and without HIV although the mechanism for this association is not fully understood.5–7
Given the number of laboratory markers that may be associated with mortality, there is a need for distinction between those abnormalities which are strongly associated with mortality and those which demonstrate a weaker association. Furthermore, the assessment of laboratory markers in those with HIV infection has generally focused on values that are above (or below) some threshold value; these thresholds are often based on those used in the general HIV-negative population (often without any further validation in those with HIV) and/or may have been chosen simply for clinical convenience. In this study, we assessed the association with risk of mortality for several biomarkers measured in people with HIV receiving antiretroviral treatment and use these markers to derive a score that can be used to identify individuals at high risk of death. Our score is based on biomarker measurements taken during routine clinical care and allows us to present a detailed analysis using more than 10 years of follow-up from the UK Collaborative HIV Cohort (CHIC) Study.
The UK CHIC Study is an observational cohort of HIV-positive individuals attending some of the largest HIV clinical centers in the United Kingdom (Appendix I). Data collected include information on patient demographics, antiretroviral history, laboratory findings, AIDS-defining events, and deaths.8 Cause of death is incomplete for the majority of records and hence it is not possible to make the distinction between HIV-related and non–HIV-related deaths.
Seven of the 12 clinics that currently participate in the UK CHIC study provide data on laboratory measurements; these clinics form the study population from which patients were drawn. Eligible patients within these clinics were those who started highly active antiretroviral therapy (HAART) after the year 2000 and who had at least 1 CD4 count, viral load, and one of the following laboratory parameters recorded after the date of starting HAART as follows: ALT, ALB, amylase (AMY), ALK, BIL, total cholesterol (CHL), CRE, glucose (GLU), hemoglobin (HGB), triglycerides (TRI), and URE. Guidelines state that these markers should be routinely monitored after starting HAART.9
Differences between those who had at least 1 laboratory marker measured after the start of HAART and those who had no laboratory measurements were tested for statistical significance using χ2 and Mann–Whitney tests. The proportion of patients with laboratory measurements within the first 3 months of starting HAART (baseline measurement), and at any time after starting HAART were calculated.
The associations between CD4 counts, viral loads, and mortality is well recognized, and hence patients were followed from the start date of HAART if they had both CD4 and viral load measurements before this date or from the earliest date at which both CD4 and viral load measurements were available after starting HAART. Patients were followed to the date of death or last visit for those who did not die over the course of follow-up. All laboratory measurements were updated in the analysis when a new measurement was made available (a time-dependent analysis); thus, reported associations reflect risk estimates over the short term (ie, the usual time interval between consecutive measurements when performed under routine clinical conditions). Associations between each laboratory marker (stratified into quintiles) and mortality were assessed using Poisson regression. In sensitivity analyses, the date of the biomarkers was lagged by 6 months. This was to ensure that the values of the biomarkers were the potential cause rather than the effect of mortality. Results were very similar to those in the main analyses (data not shown).
If a linear trend was evident in these univariable analyses, the variables were fitted as continuous covariates in multivariable analyses. Analyses were also adjusted for demographics (age, ethnicity, exposure group, sex), treatment duration and regimen, current CD4 count, current viral load and year of starting HAART. The estimates of the final model were used to construct a score for predicting short-term mortality.
In the initial analyses described above, patients with missing laboratory measurements were assigned to the quintile representing a normal laboratory measurement (either the lowest or highest quintile, as appropriate). Thus, we made the assumption that the individual's measurement, had they had one, would have been in the normal range. However, to further investigate the impact of any missing data, sensitivity analyses were performed in which individuals who had missing data were excluded, that is, patients were only included in the analyses when all laboratory measurements were available. In further analyses, baseline measurements were imputed for those with missing laboratory measurements using multiple imputation (PROC MI in SAS). Variables included in the model used to impute the missing values were sex, ethnicity, exposure, age at start of follow-up, CD4 count at start of follow up, viral load at start of follow up, year of starting HAART, and whether or not the patient died. Five datasets were imputed and the MIANALYSE procedure in SAS was used to combine the relative rate estimates from the five different datasets.
Finally, we used methods similar to those presented by the Swiss HIV Cohort Study (SHCS),10 to develop an alternative score. In particular, abnormal values of all markers were graded according to severity using the Table for Grading Severity of Adult Adverse Experiences,11 where grades 1 through to 4 reflect mild, moderate, severe, and potentially life-threatening severity, respectively. Of note, the score previously reported by the SHCS was based on a fixed-covariate analysis with measurements taken at baseline only. Thus, the SHCS score provides an estimate of the long-term risk associated with the measurements, in contrast to the short-term risk estimate provided by our own scores. As with the SHCS score, the grades for HGB were reclassified to allow for lesser degrees of anemia, which have been associated with decreased survival. Scores were assigned to each grade of abnormality; 1 point was attributed for each mild abnormality, 2 for each moderate abnormality, 3 for each severe abnormality, and 4 for each potentially life-threatening abnormality. Laboratory markers, which did not meet the criteria defined as “mild”, and those with missing data were given a score of 0. Associations between this overall score and mortality were assessed using Poisson regression and adjusted for factors mentioned above.
The Akaike's Information Criteria (AIC) was used to compare the goodness of fit of the various models, with the C-statistic used to compare the ability of each model to discriminate between those who did and did not die. The preferred model is that with the lowest AIC and higher C-statistic.
Patients and Availability of Laboratory Measures
Of the 8646 patients who started HAART at 1 of the 7 participating centers in or after the year 2000, 7232 (83.6%) patients had at least 1 laboratory measurement recorded after starting HAART. Patients without any laboratory measurements recorded after starting HAART were more likely to be in the “other” (ie, those infected nonsexually) exposure group (2.7% of those in the “other” exposure group had no laboratory measurements compared with 2.3% of men having sex with men (MSM) and 1.6% of heterosexuals, P = 0.06) and had started HAART in later calendar years (3.7% of those who started HAART in 2006–2008 had no laboratory measurements recorded compared with 0.9% of those who started HAART in 2000–2002 and 2.0% of those who started HAART in 2003–2005, P < 0.0001). Patients without laboratory measurements recorded also had lower viral loads at the start of HAART (4.5 log copies/mL compared with 4.8 log copies/mL amongst those with laboratory measurements recorded, P = 0.005). Differences between the 2 groups in sex, ethnicity, HAART regimen initiated, CD4 count at start of HAART, and age at start of HAART were not statistically significant. Baseline characteristics of the 7232 patients included in the analyses are shown in Table 1.
Table 2 shows the proportion of patients with each laboratory marker measured after starting HAART. More than 80% of patients had ALT, ALB, ALK, BIL, CHL, CRE, HGB, TRI, and URE measured at least once after starting HAART, although more than 70% of patients had at least 1 measurement of GLU, HDL, and LDL. Fewer than 60% of patients had AMY measured. The proportion of patients with a laboratory measurement within the first 3 months of starting HAART was lower, with fewer than 50% of patients having AMY, HDL, or LDL measured in this period. For this reason, together with the strong correlation between HDL and TRI, and between CHL and LDL, HDL, LDL, and AMY were not included in further analyses. The median number of measurements over the course of follow-up was more than 10 for ALT, ALB, ALK, BIL, CRE, and HGB and was 7 for CHL, TRI, and URE. The median number of GLU measurements over the course of follow-up was slightly lower (median = 5).
Evaluation of Score Based on Raw Biomarker Values
Missing Measurements Assumed to be “Normal”
In total, 247 deaths occurred over 24,796 person-years of follow-up, giving an overall mortality rate of 1.00 [95% confidence interval (CI): 0.87 to 1.12]/100 person-years of follow-up. In univariable analyses, ALT, ALB, ALK, CHL, HGB, and URE were significantly associated with mortality (Table 3). ALB and HGB both seemed to display a linear association with mortality, with lower levels of each being associated with a higher risk; for subsequent analyses, therefore, these markers were treated as continuous covariates. In contrast, associations between mortality and ALT/ALK levels seemed to be nonlinear, with a higher risk of mortality seen in those with very high values of these markers and in those with very low values. Thus, subsequent multivariable analyses retained these as categorical covariates. Finally, although significant in univariable analyses, CHL and URE were no longer significantly associated with mortality after adjusting for other variables in the model and were therefore excluded from the final model. Thus, the final model included 4 following laboratory markers: ALT, ALB, ALK, and HGB. Estimates from the final model (which can be derived using the exponential of the rate ratios (RRs) shown in Table 3) were used to construct a score to predict mortality (Appendix I). The AIC for this model was 3175, and the C-statistic was 0.78. Other factors predictive of mortality included male sex, white ethnicity, and MSM exposure. Individuals of older age, lower current CD4 counts, and higher current viral load were also at an increased risk of mortality.
Complete Case Analysis
After excluding all patients with any missing laboratory markers, the mortality rate was lower than that seen in the main analysis: 61/6736 = 0.91 (0.68 to 1.13) per 100 person-years. ALT, ALB, ALK, BIL, CHL, HGB, and URE were all significantly associated with mortality in univariable analyses. However, only 3 of these markers, ALK, ALB, and HGB, remained significantly associated with mortality in multivariable analyses. The AIC was considerably lower than that seen in the main analysis (871 compared with 3175, partly due to the lower number of patients included in this analysis), implying that this was the model of best fit. However, the C-statistic was lower than that seen in the main analysis (C = 0.76), and person-time represented in this model was only 27% of that in the main model, that is, the majority of patients were excluded from this model due to missing data.
Missing Measurements Imputed
In further analyses, baseline measurements were imputed for those with missing measurements. In univariable analyses, in addition to the variables found to be significantly associated with mortality in the main analysis, CRE was also significantly associated with mortality. ALT was not significantly associated with mortality in multivariable analyses, and hence was excluded from the final model. The other 6 laboratory markers (ALB, ALK, CHL, CRE, HGB, and URE) remained significantly associated with mortality in multivariable analyses. However, despite the higher number of biomarkers included in this model, the C-statistic was identical to the one in main analysis (C = 0.78), and the AIC was considerably higher in the models using imputed data than that in the main analysis (>236 units difference). Hence this model was not considered a better fit than that seen in the main analysis.
Evaluation of Score Based on the Grading Severity of Adult Adverse Experiences Tables
Finally, mortality rates were calculated in the different strata for each laboratory marker as utilized by the SHCS. An increasing trend between the rate of mortality and the laboratory marker severity grade was particularly evident for ALK, ALB, and HGB (data not shown). A clear association was also seen between the total score (calculated by summing up individual grades for each laboratory marker) and mortality: rates for a total score of 0, 1, 2, 3 and ≥4 were 0.27 (0.16 to 0.39), 0.69 (0.50 to 0.88), 0.88 (0.60 to 1.16), 0.93 (0.55 to 1.31) and 4.50 (3.67 to 5.33) per 100 person-years, respectively.
The total score was significantly associated with mortality in unadjusted analyses (Table 4). Patients with a total score of ≥4 were at a 16-fold increased risk of experiencing mortality compared with those with a total score of 0 [RR: 16.34 (95% CI: 10.44 to 25.58)]. After adjusting for potential confounding variables, the total score remained significantly associated with mortality. A trend in risk over categories was evident; RRs (95% CIs) for patients with scores of 1, 2, 3, and ≥4 were 2.40 (1.46 to 3.94), 2.73 (1.61 to 4.61), 2.88 (1.59 to 5.11), and 11.29 (7.12 to 19.91), respectively, compared with those with a total score of 0. The AIC for this model was 3512, and the C-statistic was 0.74.
We have identified a strong association between 4 biomarkers measured to monitor safety in people on ART (ALT, ALB, ALK, and HGB) and short-term mortality. These markers, along with demographic variables, HAART regimen, current CD4 count, and current viral load have been used to construct a score to predict mortality. Importantly, we have shown that it is not just extreme levels of these markers which are associated with mortality—a one unit increment in ALB is associated with a 4% lower risk of mortality, although a 1 unit increment in hemoglobin is associated with a 19% lower risk. We have also shown that as well as very high levels of ALT, very low levels of ALT are associated with mortality. One explanation for this is that ALT levels are known to decrease with older age.12 However, our final model was adjusted for age, and the association between low ALT and mortality remained significant.
In complete case analysis, only ALB, ALK, and HGB were retained in the final model. The C-statistic was lower than that seen in the main analysis (0.76 vs. 0.78), and estimates were weaker than the main analysis. This was somewhat surprising, since it may be assumed that patients with complete data are those who are more inclined to have abnormal measurements and hence are monitored more regularly. If this is the case, then the associations in the main analyses are likely to be marginally higher than that in real practice. However, it may also be true that those with complete measurements are no different than those without. This score was similar to the original analysis in terms of the biomarkers retained, suggesting that the assumption made previously regarding assigning measurements within the normal range for those with missing data was acceptable. However, the majority of patients were excluded from this analysis and results were therefore deemed to be potentially biased.
We also performed analyses in which missing values were imputed. Hence a full dataset was available for all patients included in the initial analyses. This resulted in a higher number of biomarkers contributing to the overall score—compared with the selected biomarkers described above, ALT was excluded, whereas CHL, CRE, and URE were included. The C-statistic of this model was identical to that of the main model, despite the higher number of biomarkers, suggesting that ability of the model to discriminate between those who did and did not die had not been improved by adding additional biomarkers.
In further analyses, we derived a scoring system for adverse events using methods similar to those used by the SHCS10 4 years ago. The scoring system derived by the SHCS was based on a fixed-covariate analysis with measurements taken at baseline only and hence the effect on mortality of new adverse events that arose over time could not be assessed. Our score derived using this approach confirmed results reported from the SHCS, in that patients with a higher total score were at a significantly increased risk of mortality. We identified a strong association between a composite biomarker score of safety measures and risk of mortality – this was again mainly driven by ALK, HGB, and ALB. Although all these have been shown to be associated with death, it is useful to provide a score to better enable assessment of the risk of mortality. We have shown that a score based on the raw values of the biomarkers discriminates better between those who did and did not die than a score similar to the SHCS score (C = 0.78 vs. 0.74). Hence it is important to recognize that a small change in some biomarkers, even within the normal range, should be considered as clinically relevant.
The association between abnormal values of these particular laboratory markers and morbidity and/or mortality in HIV-positive patients is well recognized. For example, high ALT levels suggest liver tissue damage, whereas low ALB has been shown to be associated with mortality.5–7 ALB and ALK have also been used to measure liver function and patients receiving drugs such as NVP, d4T, and ddI have been shown to be at an increased risk of liver injury.13–15 TDF has been associated with elevated ALK.16,17 Abnormalities in ALK may signify obstructed bile flow and liver complications. The association between ZDV and anemia has been well documented,18–20 and even when used as part of combination therapy, ZDV use has been shown to be a risk factor for anemia.21 DdI and d4T have also been associated with low hemoglobin levels, though not as frequently or as strongly as ZDV.22,23 Low HGB is, in turn, associated with an increased risk of mortality,5,24 though it has also been recognized that unwell individuals have lower levels of HGB. For this reason, a sensitivity analysis was performed in which the biomarker measurements were lagged by 6 months and results were in line with the main analysis.
Given the number of laboratory markers that may be associated with mortality, it is important that these markers are measured regularly and results of one marker are interpreted after taking into account results of other markers. Hence a score used to predict mortality, which combines the results of these markers may be particularly useful in clinical practice. This score will enable clinicians to identify patients who are at higher mortality risk and who may require more frequent monitoring. For these patients, an assessment of biomarker changes over subsequent visits may contribute to the overall clinical assessment of the patient. A similar technique has been used by Justice et al25 to identify a single index incorporating non-HIV markers such as HGB, eGFR, and fibrosis score; in a study population of 5980 individuals, this index added discrimination to a model which did not initially include these markers (C-statistic improved from 0.68 to 0.72). In our analyses, we have shown that the score derived using the raw biomarkers had a higher C-statistic than that derived by using methods similar to the SHCS, suggesting that the former model was better able to discriminate between those who did and did not die.
Most scores have traditionally utilized threshold values that often purport to be “clinically relevant”, with patients being designated at high or low risk if their biomarker value exceeds or falls below a threshold. These thresholds have traditionally been favored by clinicians because of their ease of use. However, the thresholds used are often based on those used in the HIV-negative population, and have rarely been validated in the HIV-positive population, where causes of mortality and morbidity will differ. Furthermore, the use of scores based on threshold values stems from an era where clinicians had limited access to technology at the bedside. Given the technological advances that have occurred over the past decade, this is no longer a key requirement. Most importantly, our results suggest that by using such a score, some individuals at high risk of mortality may not be identified promptly.
In routine care, assigning a score to those with missing measurements is problematic. Patients who have reported no adverse effects may be less likely to have a laboratory test performed, and although in most cases it is likely that their laboratory values are in the normal range, this may not always be the case. Restricting analyses to only those with all laboratory measurements available reduces the number of patients in the analysis considerably. This restriction may also result in bias because patients with all laboratory measurements recorded are likely to have different characteristics compared with those with missing measurements.
We took several approaches to deal with missing data and showed that despite which method was used, biomarkers such as ALB, ALK, and HGB were predictive of mortality.
Deriving a score to accurately predict mortality is difficult, given the many variables that could potentially contribute to the score and the diversity of the causes of death. The score must take into account both clinical and statistical issues; in a real-life setting where it is unlikely that a patient will have all laboratory measurements available, it may be impractical to have a score which relies on a full laboratory medical record. Use of measures at start of ART to predict mortality is also difficult, since a value measured at start of HAART may not be an accurate measure of the risk of mortality in years to come.
Our analyses do have the limitations inherent with observational cohort studies. Models may not have been adjusted for other confounders, such as alcohol consumption, which may impact both on the score and mortality. We were also unable to adjust for current hepatitis status which may have explained abnormal liver function tests. Further, our proposed score has not been validated. We recognize that it was possible to use internal validation methods, such as cross validation, to validate our score. However, by nature, the score will predict well on the dataset used to derive it and results of such validation may be overly optimistic. Hence, we feel that external validation is most appropriate, and we are seeking to identify an independent dataset for this purpose.
It is important to recognize the risk of mortality amongst patients with laboratory measurements in the normal range and hence we suggest using a scoring system which relies on the raw values of these laboratory markers. In the score derived from categorizing laboratory markers according to severity, ALB, ALK, and HGB were the 3 main markers influencing the score. We found the same markers, together with ALT, to be associated with mortality when focusing on the raw values of these markers. All 4 markers are measured routinely and, further, these variables have been associated with drug-related toxicities.
Using only 4 markers with a known association with mortality will limit the issues surrounding bias due to variable selection and missing data and using only the latest ALT, ALB, ALK, and HGB values is both practical and efficient, both clinically and statistically.
These data have been set up in months, and therefore, the equation above gives a monthly probability of mortality.
A white MSM individual of 38 years of age, who has a current CD4 count of 500, a VL of <50, currently receiving an nonnucleoside reverse transcriptase inhibitor–based regimen, and has the following laboratory measurements: ALT (60), ALB (30), ALK (63), and HGB (13) has the following probability of dying:
1. May M. Impact on life expectancy of late diagnosis and treatment of HIV
-1 infected individuals: UK CHIC. Presented at: Tenth International Congress on Drug Therapy in HIV
Infection; 2010; Glasgow, United Kingdom.
2. Palella FJ Jr, Baker RK, Moorman AC, et al.. Mortality
in the highly active antiretroviral therapy era: changing causes of death and disease in the HIV
outpatient study. J Acquir Immune Defic Syndr. 2006;43:27–34.
3. Krentz HB, Kliewer G, Gill MJ. Changing mortality
rates and causes of death for HIV
-infected individuals living in Southern Alberta, Canada from 1984 to 2003. HIV
4. Friis-Moller N, Sabin CA, Weber R, et al.. Combination antiretroviral therapy and the risk of myocardial infarction. N Engl J Med. 2003;349:1993–2003.
5. Shah S, Smith C, Lampe F, et al.. Haemoglobin and albumin as markers of HIV
disease progression in the highly active antiretrovial therapy era: relationships with gender. HIV
6. Sabin C, Griffioen A, Yee T, et al.. Markers of HIV
-1 disease progression in individuals with haemophilia coinfected with hepatitis C virus: a longitudinal study. Lancet. 2002;360:1546–1551.
7. Phillips A, Shaper AG, Whincup PH. Association between serum albumin and mortality
from cardiovascular disease, cancer, and other causes. Lancet. 1989;2:1434–1436.
8. Uk CHIC SC. The creation of a large UK-based multicentre cohort of HIV
-infected individuals: The UK Collaborative HIV
Cohort (UK CHIC) Study. HIV
9. Asboe D, Aitken C, Boffito M, et al.. Routine investigation and monitoring of HIV
-infected individuals. BHIVA. 2010. Available at: http://www.bhiva.org/documents/Guidelines/Monitoring/BHIVA_Monitoring_Guidelines_Consultation_version_06_September.pdf
. Accessed March 28, 2012.
10. Keiser O, Fellay J, Opravil M, et al.. Adverse events to antiretrovirals in the Swiss HIV
Cohort Study: effect on mortality
and treatment modification. Antivir Ther. 2007;12:1157–1164.
11. AIDS Clinical Trials Group. Table for grading severity of adult adverse experiences: Division of AIDS, National Institute of Allergy and Infectious Diseases, 1996. 2006.
12. Elinav E, Ben-Dov IZ, Ackerman E, et al.. Correlation between serum alanine aminotransferase activity and age: an inverted U curve pattern. Am J Gastroenterol. 2005;100:2201–2204.
13. Gisolf E, Dreezen C, Danner S, et al.. and Prometheus Study Group. Risk factors for hepatotoxicity in HIV
-1-infected patients receiving ritonavir and saquinavir with or without stavudine. Prometheus Study Group. Clin Infect Dis. 2000;31:1234–1239.
14. Wit F, Weverling G, Weel J, et al.. Incidence of and risk factors for severe hepatotoxicity associated with antiretroviral combination therapy. J Infect Dis. 2002;186:23–31.
15. Requena D, Nunez M, Nacher J, et al.. Liver toxicity caused by nevirapine. AIDS. 2002;16:290–291.
16. Welz T, Childs K, Ibrahim F, et al.. Efavirenz is associated with severe vitamin D deficiency and increased alkaline phosphatase. AIDS. 2010;24:1923–1928.
17. Fux CA, Rauch A, Simcock M, et al.. Tenofovir use is associated with an increase in serum alkaline phosphatase in the Swiss HIV
Cohort Study. Antivir Ther. 2008;13:1077–1082.
18. Fischl M, Richman D, Causey D, et al.. Prolonged zidovudine therapy in patients with AIDS and advanced AIDS-related complex. AZT Collaborative Working Group. JAMA. 1989;262:2405–2410.
19. Simpson D. Human immunodeficiency virus-associated dementia: review of pathogenesis, prophylaxis, and treatment studies of zidovudine therapy. Clin Infect Dis. 1999;29:19–34.
20. Bain B. Pathogenesis and pathophysiology of anemia in HIV
infection. Curr Opin Hematol. 1999;6:89–93.
21. Curkendall S, Richardson J, Emons M, et al.. Incidence of anaemia among HIV
-infected patients treated with highly active antiretroviral therapy. HIV
22. Sullivan P, Hanson D, Chu S, et al.. Epidemiology of anemia in human immunodeficiency virus (HIV
)-infected persons: results from the multistate adult and adolescent spectrum of HIV
disease surveillance project. Blood. 1998;91:301–308.
23. Weinberg R, Chusid E, Galperin Y, et al.. Effect of antiviral drugs and hematopoietic growth factors on in vitro erythropoiesis. Mt Sinai J Med. 1998;65:5–13.
24. Creagh-Kirk T, Doi P, Andrews E, et al.. Survival experience among patients with AIDS receiving zidovudine. Follow-up of patients in a compassionate plea program. JAMA. 1988;260:3009–3015.
25. Justice A, Modur S, Althoff K, et al. A prognostic index for those aging with HIV
: extension of the VACS index to those on cART. Presented at: 18th Conference on Retroviruses and Opportunistic Infections; Boston, MA, February 27 – March 2, 2011.
Estimates From Model in Which Missing Measurements are Treated as “Normal” (Logarithmic of RRs Shown in Table 3)
Members of the UK Collaborative HIV Cohort (CHIC) Study
UK CHIC Steering Committee: Jonathan Ainsworth, Jane Anderson, Abdel Babiker, Loveleen Bansi, David Chadwick, Valerie Delpech, David Dunn, Martin Fisher, Brian Gazzard, Richard Gilson, Mark Gompels, Teresa Hill, Margaret Johnson, Clifford Leen, Mark Nelson, Chloe Orkin, Adrian Palfreeman, Andrew Phillips, Deenan Pillay, Frank Post, Caroline Sabin (PI), Memory Sachikonye, Achim Schwenk, and John Walsh.
Central Co-ordination: Royal Free NHS Trust and RFUCMS, London (Loveleen Bansi, TeresaHill, Susie Huntington, Andrew Phillips, Caroline Sabin); Medical Research Council Clinical Trials Unit (MRC CTU), London (David Dunn, Adam Glabay).
Participating Centres: Barts and The London NHS Trust, London (C. Orkin, N. Garrett, J. Lynch, J. Hand, C. de Souza); Brighton and Sussex University Hospitals NHS Trust (M. Fisher, N. Perry, S. Tilbury, D. Churchill); Chelsea and Westminster Hospital NHS Trust, London (B. Gazzard, M. Nelson, M. Waxman, D. Asboe, S. Mandalia); Health Protection Agency—Centre for Infections London (HPA) (V. Delpech); Homerton University Hospital NHS Trust, London (J. Anderson, S. Munshi); King's College Hospital NHS Foundation Trust, London (F. Post, H. Korat, C. Taylor, Z. Gleisner, F. Ibrahim, L. Campbell); Mortimer Market Centre, London (R. Gilson, N. Brima, I. Williams); North Middlesex University Hospital NHS Trust, London (A. Schwenk, J. Ainsworth, C. Wood, S. Miller); Royal Free NHS Trust and UCL Medical School, London (M. Johnson, M. Youle, F. Lampe, C. Smith, H. Grabowska, C. Chaloner, D. Puradiredja); St. Mary's Hospital, London (J. Walsh, J. Weber, F. Ramzan, N. Mackie, A. Winston); The Lothian University Hospitals NHS Trust, Edinburgh (C. Leen, A. Wilson); North Bristol NHS Trust (M. Gompels, S. Allan); University of Leicester NHS Trust (A. Palfreeman, A. Moore); South Tees Hospitals NHS Foundation Trust (D. Chadwick, K. Wakeman):