Introduction

Intra-abdominal infections are the second most frequent cause of sepsis worldwide, only surpassed by pneumonia. These infections can be classified as uncomplicated or complicated depending on their severity. Uncomplicated intra-abdominal infections seldom cause critical illness. On the contrary, complicated intra-abdominal infections, also referred to as “abdominal sepsis,” lead to localized or diffuse peritonitis and subsequent sepsis¹. Determinants of complicated infections include patient susceptibility, age, and comorbidities, among others².

Abdominal sepsis is distinguished by a systemic response to infection which initially develops within the organs of the abdominal cavity³. Up to ~5% of patients admitted to an intensive care unit (ICU) due to infectious diseases ultimately have abdominal sepsis⁴,⁵. Worldwide, it is a major cause of non-trauma deaths in emergency departments⁶ with mortality rates being as high as 40%⁷. Therefore, prompt recognition of the infection site and early antimicrobial therapy are essential to reduce mortality⁸. A control on the source of infection should occur within 6-12 h after diagnosis since each hour of delay is associated with higher mortality rates⁹.

Various scoring systems have been developed and validated to predict mortality in patients with abdominal sepsis, such as the Simplified Acute Physiology Score II (SAPSII), Modified Shock Index (MSI), and Mannheim Peritonitis Index (MPI), all of which allow for identification of patients at high mortality risk (e.g. MPI > 25 points)¹⁰. MPI is a mortality prediction index that assesses eight risk factors, with scores ranging from 0 to 47 points. Its sensitivity and specificity are deemed at 95.9% and 80%, respectively¹¹,¹².

In addition, numerous biomarkers such as procalcitonin, neutrophil-to-lymphocyte ratio, and C-reactive protein have also been assessed as potential mortality predictors. Most studies have addressed them separately, and few studies have addressed them concurrently¹³-¹⁷. On the other hand, the role of acid-base disturbances, though considered important predictors of adverse outcomes in sepsis¹⁸-²⁰ since they reflect impairment at different levels of organ systems, has seldom been explored in the specific context of intra-abdominal infections, with the exception of serum lactate²¹,²².

For the above reasons, the interaction between different scoring systems and laboratory parameters, as well as their weight on mortality in patients with abdominal sepsis, has not been extensively studied. Furthermore, laboratory tests that assess acid-base balance are vast in number, making the identification of those that better explain mortality a priority. Thus, the goal of this study was to assess the association of laboratory parameters, SAPSII, MSI, and MPI scores with in-hospital mortality among ICU patients with abdominal sepsis.

Material and methods

Results

Out of 127 potentially eligible patients identified during the entire study period, 60 patients who met inclusion criteria were included for analyses (Fig. 1). Patient characteristics are shown in Table 1. Thirty-seven patients (61.7%) were female and mean age was 57.7 (SD: 16.7) years. Most patients were located in the > 60 years category (n = 29, 48.3%). HTN was the most frequent comorbidity (n = 27, 45%). Forty-four (73.3%) patients had a pH < 7.35 and 46 (76.7%) had a serum Cr > 1.2 mg/dl. Fifty (83.3%) patients had a MSI > 1.3 and 36 (60%) had an MPI score < 25 at admission. The most frequent SAPSII category was 64–77 points (n = 44, 73.3%). The mortality rate in the sample was 48.3% (n = 29). The non-survivor group had a higher frequency of HTN, higher lactate levels, and higher MPI scored, but lower BE and alactic BE levels. Comparison of characteristics of survivors against non-survivors is shown in table 1.

Figure 1 Patient inclusion flowchart and reasons for exclusion. 

Correlation coefficients among laboratory variables are shown in Figure 2. Highly positive correlation occurred between SIDa and SIG (p = 0.96), AG and cAG (p = 0.97), CaO₂ and Hb (p = 0.98), BE and alactic BE (p = 0.96), and PaCO₂ and PvCO₂ (p = 0.91), whereas perfect negative correlation was solely noted between SIDe and albumin (−1). The anti-image correlation matrix shows KMO values > 0.5 in all variables, which implies that all of the latter are suitable for PCA (Supplementary Table 4).

Figure 2 Heat map. Red color indicates a positive correlation between variables, while blue color indicates a negative correlation. 

PCA retrieved 8 PC with the following eigenvalues: PC1 = 5.4, PC2 = 3.6, PC3 = 2.9, PC4 = 2.8, PC5 = 2.3, PC6 = 2.1, PC7 = 1.9, PC8 = 1.2 (Supplementary Table 5), and explaining: 20%, 13%, 11%, 10%, 9%, 8%, 7%, and 5% of the total variance, respectively. Altogether, the 8 PC explain 83% of the total variance, with a cumulative proportion of 93%. Table S6 shows laboratory variables that display high loading for PC: PC1: pH (−0.59), AG (0.9), cAG (0.92), alactic BE (−0.76), HCO₃ (−0.77), and BE (−0.78); PC2: PaO₂ (−0.46), lactate (−0.45), CaO₂ (0.51), SIDa (0.60), SIG (0.67), Hb (0.58), and Cl− (−0.58); PC3: PvO₂ (0.40), PaCO₂ (0.69), CvO₂ (0.68), and PCT (0.51); PC4: SIDe (−0.59), Na+ (0.56), Alb (0.59), Cr (−0.59), and Ur (−0.51); PC5: ScvO₂ (0.69) and DvaCO₂ (−0.47); PC6: PvCO₂ (0.63); and PC7: SaO₂ (0.45) and BNP (0.48). Table S6 also shows that variables PaO₂, PvO₂, PCT, Cr, ScvO₂, Ur, PvCO₂, SaO_2, and BNP contributed the least in terms of variation to factor analysis since they showed extraction values of 0.78, 0.85, 0.80, 0.84, 0.88, 0.55, 0.85, 0.82, and 0.77, respectively.

On logistic regression univariate analyses, MPI (OR = 9.87, 95% CI: 3.07-36.61, p = 0.0002), SAPSII (OR = 1.07, 95% CI: 1.01-1.14, p = 0.01), and PC1 (OR = 2.13, 95% CI: 1.12-4.76, p = 0.04) were significantly associated with patient in-hospital mortality (Table 2), whereas only MPI (OR = 10.1, 95% CI: 3.03-40.06, p = 0.0003) and SAPSII (OR = 1.07, 95% CI 1.01-1.14, p = 0.02) were significantly associated with mortality after adjusting for age and sex (Table 3). Both models MPI (Hosmer and Lemeshow = 0.40, AUC = 0.79, AIC = 73.37) and SAPSII (Hosmer and Lemeshow = 0.70, AUC = 0.70, AIC = 82.9) showed higher performance in comparison to PC1 (Hosmer and Lemeshow = 0.33, AUC = 0.66, AIC = 84.82). Model assumptions for MPI and SAPSII are shown in figures 3 and 4, respectively.

Figure 3 Model assumptions for logistic regression analysis of MPI: (A) influential values by Cook's distance values for the top 3 largest values and (B) standardized residuals values. 

Figure 4 Model assumptions for logistic regression analysis of SAPSII: (A) influential values by Cook's distance values for the top 3 largest values and (B) standardized residuals values. 

Discussion

We hypothesized that the scoring systems and laboratory parameters have different associations with in-hospital mortality in patients with abdominal sepsis. In logistic regression models, MPI, SAPSII, and PC1 were significantly associated with in-hospital mortality. Nevertheless, after adjusting for age and sex, only MPI, and SAPSII remained significantly associated with in-hospital mortality. In addition, we found that non-survivors showed a higher frequency of HTN, higher lactate levels, and higher MPI scores but lower BE and alactic BE levels.

To the best of our knowledge, this is the first study to simultaneously evaluate various scoring systems and biomarkers in abdominal sepsis patients. Our findings could suggest that initial and follow-up assessments of patients be based on the markers that were indeed associated with mortality. In that manner, the choice of a handful of relevant parameters could pave the way for a personalized medical approach in patients with abdominal sepsis. Nonetheless, it is also important to note that our findings are merely observational; hence, they should be viewed as hypotheses generators for future clinical studies, in which relevant markers are assessed as potential therapeutic targets.

Although PC1 – a linear combination of variables pH, AG, cAG, alactic BE, HCO₃^, and BE – was not statistically significantly associated with the outcome after adjusting for confounders, acid-base disturbances still play an important role in abdominal sepsis-associated mortality and their prognostic value should not be underestimated. Acidemia (pH < 7.35) remains a potentially fatal condition, mainly due to its effect on the cardiovascular system. The classification of acidemia on various subtypes such as metabolic (BE < −2 mEq/L and PaCO₂ ≤ 42 mmHg), respiratory (BE ≥ −2 mEq/L and PaCO₂ > 42 mmHg), and mixed (BE < −2 mEq/L and PaCO₂ > 42 mmHg) acidemia is of paramount importance when determining appropriate treatment and prognosis. In fact, mortality in metabolic acidemia is as high as 11% when BE is < 6.7 mEq/L; in respiratory acidemia, it is of 6% when PaCO₂ is > 51 mmHg, and in mixed acidemia, it is of 13% when pH is < 7.23²⁸. A persistent state of metabolic acidosis harbors negative consequences on patient outcomes and it may also indicate progression of underlying causes. A standard base deficit clearance < 11% (within the first 24 h of ICU admission) increases the risk of mortality in patients with septic shock (< 11% = 65.1% mortality and > 11% = 36% mortality)²⁹. A negative value of alactic BE indicates worsening of the renal function and, therefore, an impairment on the clearance of unmeasured anions and lactate. Thus, the more negative the alactic BE is, the greater the renal impairment and, most likely, the worse the prognosis³⁰. Increased AG occurs mainly due to accumulation of unmeasured anions. Its highest utility is achieved when used on the differential diagnosis of metabolic acidosis, although it bears prognostic value as well. A recent study found that the greater the AG, the greater the mortality risk: AG ≥ 13 mEq/L (OR 1.02, 95% CI 0.75-1.40, p = 0.891), AG ≥ 15 mEq/L (OR 1.40, 95% CI 1.03-1.91, p = 0.031), and AG ≥ 17 mEq/L (OR 2.78, 95% CI 2.12-3.63, p < 0.001)³¹.

Even though PCT helps predict the occurrence of intra-abdominal infections after definitive surgery for intestinal fistulae³², infected necrotizing pancreatitis³³, and it is also an early biomarker of intra-abdominal infections in post-operative patients with gastrointestinal malignancies³⁴, this biomarker might not be as useful when used as a predictor of mortality in patients with abdominal sepsis, based on our findings.

We would like to draw attention to the fact that out of all the laboratory parameters included in PC1, HCO₃ is also coincidentally an item in one of the scores associated with mortality (SAPSII). Although the statistical significance of PC1 was lost during multivariate analysis, this finding suggests that there might be room for including new laboratory parameters into the existing scoring systems that could further improve mortality risk estimates³⁵,³⁶.

In accordance with their AIC values, we conclude that MPI and SAPSII scoring systems are both equally associated with mortality. Differences observed in OR values are due to the taxonomy of the variables themselves, but the strength of association with mortality is not any different among them (Table 3). In our study, we considered MPI cutoff values as reported on the original paper³⁷, proving significant association with in-hospital mortality, which is consistent with the literature³⁸. Nonetheless, other studies have used cutoff values of > 21, while displaying sensitivity of 97.5% and AUC of 0.976.³⁹ In addition, SAPSII showed optimal performance, in similar fashion to that reported by Sánchez-Casado et al.⁴⁰ Besides association with mortality, SAPSII has also been associated with overall length of hospital stay⁴¹.

Finally, we want to emphasize that our findings are consistent with the literature regarding the relationship between HTN and adverse outcomes in sepsis patients since non-survivors showed a higher frequency of HTN. A recent study by Garg et al. showed that male gender, HTN, SOFA > 9, and increased creatinine were predictors for mortality in adult patients with diagnosis of sepsis requiring ICU care⁴². Moreover, it appears that HTN, along with other chronic medical conditions, might increase the risk of incident sepsis episodes⁴³.

Limitations of our study include its retrospective nature, as well as potential sampling bias due to convenience sampling, and having used a selective study cohort, which limits generalizability of results. Since this study was developed in a low-resource hospital in Mexico, we were unable to assess important predictors which were unavailable (i.e., C-reactive protein, interleukin-6), whereas other variables were not presented since they are already included in the scores (i.e., leukocytes and bilirubin). In addition, we did not evaluate specific causes of abdominal sepsis. Another limitation is that even when the sample size was calculated to encounter differences in the main outcome according to MPI scores, it is possible that the study was underpowered to detect differences according to PCA groups. Furthermore, PCA is a technique for reducing the dimensionality of large datasets which can be challenging to interpret; for instance, it is difficult to acknowledge the most important characteristics of the dataset after calculating PC. Although reduction of dimensionality is useful, it does not come without a cost, given that the loss of information is a necessary part of this type of analysis. Balancing dimensionality reduction and loss of information are mandatory when applying PCA.

Conclusion

MPI and SAPSII are both equally associated with in-hospital mortality in abdominal sepsis patients admitted to the ICU. Acid-base disturbances play an important role in abdominal sepsis-related mortality, although they were not significantly associated with in-hospital mortality in our study when adjusting for age and sex. Larger studies are needed to confidently evaluate the association of pH, AG, cAG, alactic BE, HCO₃, and BE with mortality in these patients to better understand pathophysiology of abdominal sepsis.

Supplementary data

Supplementary data are available at 10.24875/CIRU.22000219. These data are provided by the corresponding author and published online for the benefit of the reader. The contents of supplementary data are the sole responsibility of the authors.