The intensive care unit of the University Hospital of Parakou in Benin was used as the study setting. It is a 24-hour functional unit with 13 beds, 4 of which are equipped with multiparameter scopes with two intensive care respirators.

It was an observational study with a descriptive and analytical aim with prospective data collection carried out over a period of 4 months, from the 1^st of March 2017 to the 30^th of June 2017.

Our study population consisted of patients of more than 16 years of age who were hospitalized in intensive care for at least two (02) hours having given their free and informed consent or of the family representative. All patients younger than 16 years of age or who had been in the unit for less than two (02) hours (death, transfer) were not included in the study. An informed consent form was filled and signed by the patients or a family representative. The study was conducted after obtaining an authorization from the local ethics and biomedical research committee of the University of Parakou. An exhaustive recruitment of patients fulfilling the inclusion criteria was carried out and data were collected by the means of a questionnaire filled after reviewing medical records and conducting clinical and paraclinical exams.

The independent variables studied were sociodemographic (age, sex), clinical (type of admission, history, vegetative constants, visceral defects, Glasgow score), paraclinical (blood count and formula, azotemia/creatininemia, liver transaminases, blood cramps), progressive (good or complicated) and prognostic (alive/dead).

The dependent variables were represented by the two general severity scores selected for the study (IGS II, APACHE II). Patients were classified according to disease category. Each of the scores were scored for our patients on their admission and then calculated using the software Médicalcul version 1.32 [8], which gave us score values, as well as the predicted mortality when available. These were:

­ IGS II (Simplified Severity Index II): a simplified severity assessment system created by Le Gall et al. [9]. It is a questionnaire with 12 variables (age, type of admission (medical, urgent surgical or programmed) and underlying diseases (AIDS, metastatic cancer and hematological disease)). Score scoring is done during the first 24 hours and only the most abnormal ones observed are considered for scoring. The score obtained varies from 0 to 163 points of which 116 points for the 12 variables, 26 points maximum for Glasgow, 17 points maximum for age, 30 points maximum for chronic diseases.

­ The APACHE II (Acute Physiology and Chronic Health Evaluation II) score was developed in 1985 on the basis of 12 variables evaluated during the first 24 hours of hospitalisation in intensive care [10]. Each variable is evaluated and quoted with a value ranging from 0 (normality zone) to 4 (most abnormal value). Biological data impossible to mesure (arterial pH) which doesn’t prevent the score from being scored; a 3% margin of error set by the authors of the score for each of the missing variables and considered normal.

The qualitative variables were expressed in terms of frequency and percentage. Quantitative variables were expressed in terms of mean plus or minus standard deviation from mean. For each of the IGS II and APACHE II scores, the Receiver Operating Caracteristic (ROC) analysis allowed to split patients into two severity groups from the threshold value obtained. From the determined thresholds, patients were classified into two groups: severe and less severe state. The calibration of each score was evaluated using decision matrix analysis (threshold values set a priori) to compare observed and predicted results to a decision criterion of 1 (a calibration greater or less than 1 was bad; meanwhile, it was excellent if equal to 1). By the way, the strength of mortality prediction of each score was determined according to the indices obtained. The main variables associated with mortality in univariate analysis, including severity scores (dichotomized), were simultaneously introduced into a logistic regression model by performing successive iterations of the stepwise descending type. The strength, meaning and stability of the association were estimated using Odds ratios with their 95% confidence interval. A p < 0.05 was considered as statistically significant. All collected data have been entered, processed and analyzed with the software Epi Info v7.1.1.14 and XLSTAT version 2017. The tests used were the ANOVA test, the Fisher’s exact test or the Pearson’s Chi-2 (χ²) test.

During the study period, 207 patients were admitted to intensive care. We collected 185 patients representing 89.37% of the admissions of which 63.78% were male (sex ratio: 1.76). The mean age of the patients was 38.89 ± 17.92 years with extremes of 16 and 99 years. The most represented age group patients were those aged 16 to 25 years (27.03%). The self-employed were the most represented (60%). Unemployed patients numbered 31 (16.76%), followed by pupils and students (26; 14.05%) and employees (17; 9.19%). Table 1 shows the distribution of patients according to age, sex and profession.

Medical pathologies were the main type of admission to intensive care (50.27%). Other reasons for admission were urgent surgery (45.41%) and scheduled surgery (4.32%). In our series 71 patients (38.38%) had at least one pathological history. The pathologies identified were hypertension (16.22%), surgery (12.43%), diabetes (5.41%), HIV immunosuppression (2.70%), and metastatic cancer (1.62%). The reasons for hospitalization identified were varied. Traumatic pathologies came first (26.49%). Among them, cranioencephalic trauma (TCE) was the most frequent traumatic pathology (59.18%). Table 2 shows the distribution of patients by reason for admission.

The scoring of the scores required the search for the various visceral failures. The visceral failures observed were: neurological failure (58.38%), respiratory failure (49.19%), renal failure (41.08%), hepatic failure (12.43%), hematological failure (10.81%) and cardiocirculatory failure (9.19%).

The ROC (Receiver Operating Caracteristic) analysis allowed to find the thresholds of severity of the IGS II and APACHE II scores (Figure 1 and Figure 2).

Table 3 shows the distribution of patients according to the threshold values, mean and extreme values of severity scores.

The length of stay in intensive care ranged from 04 hours to 48 days with a mean of 4.36 ± 2.2 days. 48 deaths (25.95%) were recorded, the mean age of the death patients was 46.10 ± 19.44 years. In our series 62.5% deaths had occurred within the first 48 hours of admission, they were hospitalized for a medical (50.27%) or surgical (49.73%) pathology. At least one visceral failure had been found in 97.91% of death patients.

Regarding mortality and IGS II score, 66.66% (n = 32) of deaths occurred to those with an IGSII score ≥ 40 (severe cases) versus 33.34% (16) of those with an IGS II score < 40 (light severity). For IGS II ≥ 76 there were no survivors. The average IGS II scores for the death and surviving patients were 46.04 ± 19.30 years and 23.75 ± 15.23 p = 0.00, respectively.

Regarding the APACHE II score, 72.91% (35) of the death patients had an APACHE II ≥ 19 versus 27.09% (13) of deaths on those with an APACHE II < 19. For APACHE II ≥ 35 there were no survivors. The means of the APACHE II score for the death and for the survivors were 23.60 ± 10.40 and 10.96 ± 8.35 p < 0.001, respectively.

The IGS II and APACHE II severity scores allowed to calculate the Predicted Mortality (PM) of our patients (MP-IGS II and MP-APACHE II) and the obtained results are reported in Table 4. Distribution of patients according to the Predicted Mortality by IGS II and APACHE II scores.

Table 5 shows the performance analysis of the IGS II, APACHE II severity scores.

Table 6 presents the results of the univariate analysis between IGS II, APACHE II and death.

SD**: Standard deviation.

OR: Odds ratio IC 95%: 95% confidence interval.

Intake type

In our series, we had as many medical admissions as we have surgical admissions. However, among the reasons for surgical admissions, surgical emergencies predominated (91.30%). Our results are different from those reported by Tchoua et al. in 1999 in Gabon, who reported a predominance of medical admissions (61.57%) [11] in contrast to Diouf et al. in Senegal in 2011 and Iteke et al. in Congo in 2014, who reported a predominance of surgical admissions in the same proportions of 60% [12] [13]. This high frequency of surgical emergencies found in our study could be explained by the increasing number of traffic accidents responsible for serious injuries, particularly cranioencephalic trauma, and the delay in consulting of certain patients suffering from infectious or parasitic medical pathologies which, in the absence of early diagnosis and treatment, evolve towards surgical complications, particularly peritonitis and occlusions.

Our findings are similar to those of other African authors. Indeed, Lokossou et al. reported in a study conducted in Cotonou in 2014 that 35.7% of patients had a pathological antecedent [7]. Concerning the pathologies found, high blood pressure and diabetes predominated in many studies and specifically Lokossou et al. in Cotonou, Dia et al. in Senegal, also found high blood pressure as the main pathological antecedent with respective frequencies of 54.5% and 20.2% [7] [14]. These results are corroborated by WHO’s reports who makes hypertension a public health problem [15].

In our study, more than 75% of patients had at least one visceral failure at admission. Wade et al. found that 50.4% of patients had experienced visceral failure on admission to their series in 2011 in Senegal [4]. Neurological failure was the most common in our study, followed by respiratory failure. Our results are comparable to those of Wade et al. [4] in Senegal, who found neurological failure to be the most frequent (30.5%), followed by respiratory failure (28.8%). In contrast, Tchoua et al. [11] in 1999 in Gabon found respiratory failure to be the most common in 40% of cases, followed by neurological failure (21%). The high number of cranioencephalic traumas, high blood pressure including stroke, diabetic comas and eclampsia could justify the predominance of neurological failure in our series.

Traumatic pathologies represented the main reason for hospitalization in our series and were dominated by cranioencephalic traumas. Our results are corroborated by the results of other studies around the world [16] [17]. The predominance of traumatic pathologies could be explained by the increase in the number of road accidents resulting from the non-respect of basic traffic regulations and the poor state of the road infrastructure and parking. High blood pressure emergencies constituted the second major group of reasons for hospitalization followed by severe patients admitted for post-surgery care. Similar results were found by Diouf et al. [12] in 2011 in Senegal (29.42% of cases). We can link this high rate of patients admitted for post-surgery care in our intensive care unit to the high number of surgical emergencies, particularly visceral and trauma emergencies.

Length of stay

The average length of stay in our series was relatively short. This observation is comparable to those made by some authors who reported an average length of stay ranging from 4 to 10 days [18] [19]. So, in order to reduce the risk of infections related to intensive care and the unfavourable evolution of patients after surgery, we opt for aggressive nature of drug management, in especially by double antibiotic therapy (even triple antibiotic therapy) and rigorous monitoring, elements that contribute to the fast improvement of patients’ clinical condition and their discharge from intensive care.

Mortality

The objectified mortality in our series remains high despite the progress made in patient management compared to those reported in countries with a high technical plateau. In France, Jungfer et al. reported 18% of mortality in 2008 [20]. Grégoire et al. in Belgium and Sik et al. in Tunisia had respectively found a mortality rate of 14.7% and 16% in their series in 2017 [21] [22]. On the other hand, our mortality rate appears to be lower compared to other African series. Diouf et al. in Senegal in 2011, Iteke et al. in Congo in 2014, and Lokossou et al. in Cotonou in 2014 had objectified mortality rates of 32.7%, 32.7% and 60.5% respectively [7] [12] [13]. In general, death rates in intensive care units are higher than in other services worldwide. This is explained by the severity of the clinical condition of patients on admission, often in relation to the multivisceral failures linked to the long consultation time limit.

The IGS II system

Mortality increased significantly with a high IGS II score p < 0.001. These results are similar to those reported by other authors [23] [24]. Among the death patients, 60% had an IGS II score ≤ 40 (light severity) against 13.33% of deaths in severe patients (IGS II > 40) p < 0.001. There was a significant statistical difference between the mean IGS II values among survivors (23.75 ± 15.23) and those who died (46.04 ± 19.30) p < 0.001. An APACHE II score ≥ 19 was significantly associated with death OR = 11.5 [5.37; 24.64] p = 10⁻⁵. An IGS II score > 40 was a risk factor significantly associated with death in both univariate and multivariate analysis by logistic regression with odd ratio and P value respective of OR = 9.75 [4.59; 20.69] p = 10⁻⁵, adjusted OR = 3.22 [1.20; 8.61] p = 0.0198. The IGS II score had a standardized report of overall mortality (SMR) of 1.35 (19.12% predicted mortality for 25.95% observed mortality); indicating a poor overall adjustment of the score in our study. Deaths were relatively well estimated by the IGS II in patients of light severity with an SMR of 1.12 (12.03% of observed mortality for 5.27% of predicted mortality); in severe patients, on the other hand we had an overestimation of death by the score with an SMR of 2.28 (54.56% predicted mortality for 61.53% observed mortality). These results reflect a poor calibration of the IGS II score in our study, resulting either from poor management, poor IGS II rating (by default), or simply from an inadequacy of the IGS II score for these defined severity groups. Wade et al. [4] [11] in 2011 in Senegal had also found a poor calibration of the IGS II score in their study. The area under the ROC curve had shown a good discrimination capacity of 0.82 ± 0.04 with a sensitivity of 66.70% and a specificity of 85.40%. The found Youden index (0.52) demonstrated the effectiveness of IGS II in predicting mortality in our study. Several authors also corroborated our results. This is the case of Le Gall et al. in 1993, who found an AUC-ROC of 0.86 in the initial sample of 13,152 patients who enabled the validation of the IGS II score [25].

The APACHE II score

Mortality increased significantly with a high APACHE II score p < 0.001. These observations are similar to those of literature [26] [27]. Among the death patients, 60% had an APACHE score ≥ 19 (severe patient) against 11.54% among those with an APACHE II score < 19 (light severity) p < 0.001. In our series, the APACHE II mean points were significantly different in the death as in the alive with respectively 23.60 ± 10.40 versus 10.96 ± 8.35 p = 0.00. An APACHE II score ≥ 19 was significantly associated with death OR = 11.5 [5.37; 24.64] p < 10⁻⁵. The standardized mortality ratio (SMR) of the APACHE II score showed of its good overall adjustment in our study, with a value of 1.01 (25.69% predicted mortality for 25.95% observed mortality). There was an excellent ratio for the different severity groups with SMR values of 0.99 for light severe cases (10.66% predicted mortality for 10.57% observed mortality) and 1.05 for severe cases (55.50% predicted mortality for 58.64% observed mortality). These results show a good calibration of the APACHE II system in our study and are in line with other work carried out elsewhere. The area under the ROC curve showed a good discrimination capacity of 0.83 ± 0.08 of the score. It also demonstrated its efficiency in our series with a sensitivity of 72.90%, a specificity of 80.03% and the Youden index of 0.53. APACHE II had the best discrimination as well as the best calibration and therefore has demonstrated its ability to predict mortality sufficiently well.

APACHE II asserts its validity as a predictive tool in our unit before the IGS II. We suggest extending this prospective study to a larger sample, with a view to validating this model in countries with limited resources such as ours.