We conducted a prospective cohort study between 1 ^st August 2017 and 1 ^st March 2018 at Mulago National Referral ICU, Uganda Heart Institute ICU, International Hospital Kampala ICU and Nakasero Hospital Limited ICU in Kampala city, Uganda. Baseline data on patients’ demographic variables (employment status, education level, family income, smoking, age, sex and ethnicity), admission diagnosis, CD4 T Cell counts and APACHE II scores were collected. We included adult HIV negative critically ill, APACHE II scored, medical/surgical ICU patients and excluded patients found admitted to ICU beyond 24 hours and those on immunosuppressant drugs such as steroids prior to admission. A total of 130 critically ill HIV-negative adults were enrolled into the study of which 127 participants gave written informed assent on behalf of their critically ill patients while 3 were waived of consent by the ethics committee because they had no proxies. The sample size was calculated using the formula for sample size calculation for two groups with a continuous outcome as outlined in Designing Clinical Research by Hulley et al. ¹⁰ . We aimed for power of 80%, level of significance of 95% and using mean estimates of CD4 from a study ⁶ . All study participants were followed for 28 days and end of follow up survival and mortality data was collected.

Referring to World Health Organization, we grouped CD4 levels into two; where CD4 above 500 cells/mm³ signified immune competent or normal CD4 count and those with CD4 less than 500 cells/mm³ reflecting low immunity.

The APACHE II scores and blood draws for CD4 T cell counts were performed upon admission between 8 am and 10 am. Blood sampling followed a standard laboratory practice. Approximately 3 to 5 ml of blood were collected in K3/K2 EDTA vacutainers, labeled with the patient’s identification, date and time of collection, and the name of the collecting personnel. To assess patients’ CD4 levels, BD FACSCalibur anticoagulated blood samples transported at ambient temperature (20–25°C) was stained within 48 hours of draw and then analyzed within 6 hours of staining ¹¹ . Samples were analysed from a 4-star laboratory of Makerere-Mbarara University Joint AIDS Program. Sample transport was by hand delivery and no transport was done on non-testing days. A coding manual for laboratory results was developed for broken samples, insufficient, clotted, frozen, haemolysed blood, samples not been drawn in K3/K2 EDTA vacutainers and errors in laboratory procedures.

Strict procedures for data management during the pre-analytical, analytical and post analytical phase of testing were conducted to ensure the reliable production and delivery of accurate test results. Laboratory equipment was calibrated daily and sample laboratory registers were used to record receipt of samples and the production and release of results on entry of test result form.

The collection sites maintained the test request form. Testing laboratory had reliable systems for receiving and processing result data with uniform basic data handling, storage and reporting standards. The testing laboratory maintained records of result data for defined periods, to allow repeat reporting of lost test results, as well as aggregation for monitoring and evaluation or other research purposes. The testing laboratory also ensured reliable and rapid delivery of results.

The questionnaires were cross-checked by the principal investigator (PI) to ensure completeness before leaving the study site and periodically, the PI arranged a meeting with the assistants to validate data. Computer in-built checks reinforced data completeness. Quantitative data was double-entered to ensure correctness of data entered. According to WHO guidelines, the questionnaire was translated into Luganda a local dialect and back-translated into English by K.A.M.

To address potential sources of bias, the PI and critical care nurses (research assistants) sampled the participants by drawing blood and filling the questionnaires that were retained at the study sites. The laboratory technician (research assistant) transported all samples with only a laboratory request form and did not participate in drawing blood from the patients, only K.A.M. accessed the study results and strictly 130 participants were recruited and all completed a 28-day follow-up.

This study was approved by Research and Ethics Committee of Makerere University. A waiver of requirement for consent for unconscious patients without proxies was obtained with a reference number 2017-095. Final approval was granted by Uganda National Council for Science and Technology with a reference number HS104ES.

An electronic database was created using EpiData version 3.1 to enter the raw data from the questionnaires. The data was then transferred to STATA version 14.1 for analysis. In determining the CD4 T cell counts among the study participants, we presented the mean CD4 count with its corresponding standard deviation since it was normally distributed. In addition, we presented the CD4 as a categorized variable with frequencies (and percentages) for the various cutoffs with the corresponding 95% confidence intervals of the proportions.

In order to determine the relationship between CD4 T cell counts and 28-day ICU mortality, we performed multivariate logistic regression with CD4 count as the main predictor and 28-day mortality as the outcome. Prior to performing the multivariate logistic regression models, we performed bivariate analysis and all the variables with a p-value of 0.2 or less were included in the multivariate model.

Multivariate logistic regression was performed to determine how the CD4 jointly with other variables was associated with 28-day mortality. The variables were entered into a stepwise logistic model. Significance was set at p-value of 0.05 or less. The goodness of fit of the final model was tested using the Hosmer & Lemeshow goodness of fit, testing the null hypothesis that the final model adequately fits the data.

To assess the feasibility of using CD4 T cell counts to predict 24-hour mortality, as compared to APACHE II score, we compared the area under the Receiver Operator Characteristic Curves (ROC) between CD4 and APACHE II in predicting mortality. Prior to generating the ROC, we generated the sensitivities and specificities for the different cutoffs for both CD4 count and APACHE II. The ROC was then generated with y-axis being sensitivity and the x-axis being 1-specificity.

More than half (53.9%) of the participants were recruited from MNRH followed by IHK (24.6%), NHL (19.2%) and lastly UHI (2.3%). Non-smoking self-employed black males dominated the study population at a mean age of 45.2±18.3 (mean±SD) and a family income above $1 as shown in Table 1. The major indication for admitting to ICU was postoperative high critical care requirements (46.2%), whilst the least common was urinary tract infection (UTI) (0.8%). Details are shown in Table 2. All raw data are available on OSF ¹² .

Overall 130 CD4 tests were carried out, of which 71 [54.6%, 95% CI (45.9-63.3)] were low (less than 500 cells/mm³). The mean CD4 count was 494.4±282 cells/mm³ (mean±SD), and the lowest count was 50 cells/mm³. Other details are shown in Table 3. There was no significant association in mortality outcome between those who had normal (CD4 ≥500 cells/mm³) and low (CD4 <500 cells/mm³) CD4 counts (p = 0.64). Other details are shown in Table 4.

At bivariate analysis, smoking, admitting a patient from another hospital, ICUs for hospitals MNRH, NHL and UHI had a strong statistically significant association with mortality at day 28. At multivariate analysis, abnormal CD4 count was not found to be significantly associated with mortality at day 28 in our population OR (95%) 1 (0.4–2.4) p = 0.093. Other details are shown in Table 5.

From the receiver operator characteristic curves for comparing CD4 cell count and APACHE II score in predicting mortality, the area under the curve for the two graphs was comparable (this signified that CD4 count could be as good as APACHE II score). However, both graphs demonstrated very low area under the curve (the closer to 1 the area is, the more diagnostically accurate the curve). Therefore, the data signified that both APACHE II and CD4 were not good predictors of the outcome, despite being comparable ( Figure 1).

To our knowledge, this multicenter cohort study is the first report to discuss the immune status of critically ill HIV-negative patients admitted to Ugandan ICUs using CD4 T cell count as a surrogate marker. Almost all participants were black, of African descent and non-smokers, because black Africans, who rarely smoke, dominated the study population.

Most admissions from all the four ICUs were surgical cases and those requiring high postoperative care contributed the highest number of participants while the least was due to UTI. This is because MNRH is the referral center for most critical patients and strictly to mention the trauma patients. The same happened to UHI ICU that admitted mostly surgical cases.

In our study, we found that more than half of the participants had low CD4 T cell counts This may have been caused by critical illness that led to production of cortisol. This in turn may have suppressed the production of CD4 T cells. Our findings agree with a study conducted in nine consecutive patients admitted to the ICU with sepsis in Japan, whose CD4 cells were clearly reduced below 500 cells/mm³ and remained at that level for entire 4 weeks ¹³ . These findings are also in agreement with a study conducted in HIV-negative Senegalese individuals, which found that CD4 cell counts varied in HIV-negative individuals ⁶ . Though our study population was purely HIV negative, we found that more than 50% of the participants had low CD4 cell counts, with four participants having their CD4 cell counts as low as less than 50 cells/mm³ and six participants having counts less than 200 cells/mm³, values considered to indicate AIDS in patients living with HIV. Hence critical illness alone, without HIV infection, can present a picture that resembles that of AIDS in HIV-negative critically ill patients.

We did not find a statistically significant association between CD4 T cell counts and ICU mortality at day 28 among critically ill HIV-negative patients in our population. This is consistent with a study conducted by Feeney et al., which did not find whether low CD4 T cell counts were associated with a poor prognosis ⁹ . The reason why this American study agrees with our findings could be entirely attributable to the sample size that is almost similar in both studies. However, our results contradict with other studies that have shown that septic patients with loss of CD4 T cells have a higher mortality ¹⁴ . It is also in contrast with a study conducted in 2007, which showed that low CD4 T cell counts were associated with death ¹⁴ . Our findings could be ascribed to the fact that CD4 T cells are a surrogate marker of the many immune cells. Hence, measuring CD4 alone could not yield reliable information to predict mortality. Another reason for the lack of statistical significance observed would be due to the sample size and short-term follow-up that may be were not adequate to give dependable results. It is also prudent to note that CD4 T cells were only sampled once, hence making it hard to track the exact CD4 cells at the time of the patient’s demise.

Both high APACHE II and low CD4 count could predict a 24-hour mortality in our population; however, despite being comparable, both were not good predictors of mortality. This is in line with a study conducted in 2000, where elevated APACHE II score remained a significantly negative predictor of survival at 28-day mortality ¹⁵ . It also concurs with a study conducted in 2015 that reported that the median APACHE II of 25 predicted greater than 50% mortality ⁸ . The latter leaves a benefit of doubt, as the study did not report that mortality would be 100%. However it is in contrast with a study done in 1995 that did not find any relationship between CD4 counts and APACHE II score, predicted mortality rate, or survival rate ⁹ .