Early sepsis markers in patients admitted to intensive care unit with moderate-to-severe diabetic ketoacidosis

Diabetic ketoacidosis (DKA) accounts for 4–9% of all hospital discharge summaries among diabetic patients [1, 2]. Despite a 50% drop in mortality since 1980 due to standardized protocols [3], recent studies still report a mortality rate of about 2–5%, mostly depending on the age of the patients [4, 5]. Therapeutic guidelines on DKA almost only focus on insulin administration and hydroelectrolytic supplementation [6, 7].

Discontinuation of insulin therapy and infections are the most frequent triggering factors [2, 5, 8]. Bacterial infections—urinary tract infections in the first place, followed by pneumonia—explain up to 50% of ketoacidosis cases [6, 8, 9]. In the context of DKA, bacterial infections are reported to increase both mortality [10] and length of stay [9]. Then, early detection of bacterial infections associated with adequate antibiotic treatments are key elements to improve patient outcomes. DKA itself can however mimic infections [10] and differentiation of septic from non-septic inflammatory response may be difficult. Clinical suspicion of infection can hardly be used and many patients are over-treated with antibiotics leading to inadequate treatment costs, side effects and bacteriological resistance. The development of bacteriological resistance is all the more worrying for these patients admitted for DKA knowing that 20 to 40% of them will develop a new episode of infection in the future [3, 5, 11].

To our knowledge, no study has already assessed the usefulness of sepsis markers in DKA. The constitutive signs of the systemic inflammatory response syndrome are considered to have a poor specificity [12]. Tachycardia and polypnea can be easily integrated in DKA pathophysiology. Hypothermia, fever, and white blood cell count abnormalities are usually considered when assessing septic status. However, none of those signs ever were found to be relevant to distinguish infected from non-infected patients during DKA. Procalcitonin (PCT) is actually one of the major relevant markers for the diagnosis of bacterial infections. PCT is daily used for antibiotic decisions in patients with respiratory tract infections and sepsis [13‐15]. Nevertheless, some preliminary data suggest that compared with non-diabetics, PCT positive threshold could be higher in diabetic patients, especially during hyperglycemic crisis [16, 17]. We therefore conducted a retrospective study in which we sought to investigate the diagnostic performance of different sepsis markers (including PCT) to predict bacterial infection in the first 2 days of admission in intensive care unit (ICU) for DKA.

This is the first study to assess the diagnostic performance of different sepsis markers to predict proven bacterial infection for patients with DKA, admitted in ICU. Fever and high PCT (threshold above 1.44 ng/mL at D0) at ICU admission may help to identify patients with proven bacterial infection in the context of DKA.

The only clinical marker was temperature. Presence of fever on D0 and D2 was higher in PBI episodes as reported in previous studies [10]. Nevertheless, in episodes without PBI, body temperature ranged from 32.9 °C till 38.7 °C on D0. This huge variation may be explained by thermoregulatory function impairment in diabetic patients [22]. Hypothermia (temperature < 36 °C) was equally presented in both groups. In 1978 Gale et al. [23] reported 20 patients with hypothermia during DKA and observed a high mortality rate (60%). Hypothermia was also associated with infection [24]. In our study, we neither found an increase in mortality nor an association with sepsis for hypothermic patients.

Other classical sepsis markers were also found to be inefficient in our study to differentiate PBI episodes from those without PBI. We found a high WBC level on D0 (mostly composed of neutrophil polynuclears) in episodes without PBI. Such leukocytosis, as high as 57.0 G/L, has already been reported in several case reports [25, 26]. This result leads to reconsider the usefulness of WBC to predict bacterial infection at admission. Recently, the NLCR was proposed to be a more useful diagnostic tool than other blood tests to identify patients with bacterial infection [27]. However, in our study we did not highlight any difference for this marker between both groups at admission.

PCT, a precursor of calcitonin, is generated as part of the systemic response to bacterial infections [28]. Our study emphasized the relevance of PCT to predict infection, with a good predictive value above the level of 1.44 ng/mL at D0. In febrile patients admitted in the emergency department, Hausfater et al. [29] stated that a 0.2 ng/mL cutoff value for PCT had a low Se and Sp to diagnose bacterial infections (0.77 and 0.59). Wacker et al. [30] in their meta-analysis focusing on the accuracy and clinical value of PCT for diagnosis of sepsis in critically ill patients reported a Se and Sp of 0.77 and 0.79, respectively. Sager et al. [31] recently summarize the use of PCT to guide sepsis diagnosis. For critically ill patients, bacterial infection was considered to be “likely” when PCT level was 0.5–1.0 ng/mL and to be “very likely” above 1.0 ng/mL. In our study, PCT was accurate at the admission to distinguish PBI episodes from those without PBI with a sensitivity of 0.90 and a specificity of 0.76. However, the positive threshold seems to be higher than usual in our study (optimal cutoff on D0: 1.44 ng/mL). Previous studies already reported huge cutoff heterogeneity. For example, Wacker et al. [30] in their meta-analysis reported a median cutoff of 1.1 ng/mL (IQR 0.5–2.0 ng/mL). However, a participation of the hyperglycemic crisis in the increase of the PCT could not be excluded. Indeed, Aksu et al. [32] had reported a decrease in PCT level following a normalization of glycemia in patients with acute hyperglycemic crisis. In our study, we found an elevated a high level of PCT in patients without any proven bacterial infection, with a PCT drop following normalization of glycemia. High PCT levels were recently reported in different case reports or case series focusing on diabetes ketoacidosis without infection [33, 34]. A case series of 5 patients hospitalized for diabetes ketoacidosis reported PCT levels ranging from 6.87 to 30.47 ng/mL. Interestingly, this observation was not found in case of hyperosmolar hyperglycemic syndrome. This led the author to conclude that the augmentation of PCT in acute glycemic crisis may only be found during diabetes ketoacidosis [34].

In early management of DKA, traditional clinical (hypothermia) and biological (WBC, NLCR) signs of bacterial infection proved to be ineffective, probably because of the reported correlation between hyperglycemia crisis and inflammatory response. In non-diabetic patients, induced hyperglycemia led to an amplification of interleukin-6 (IL-6) and other pro-inflammatory markers [35]. Adding low doses of insulin avoids these alterations even with persistent hyperglycemia [35]. Compared with healthy controls, an induced hyperglycemia in diabetic patients results in a more pronounced secretion of pro-inflammatory cytokines such as tumor necrosis factor-α (TNF-α) and IL-6 [36]. Apart from any bacterial infection, TNF-α is known to induce the release of large amount of PCT in both animals [37] and humans [38]. These data may explain the increase of both PCT and WBC in episodes without PBI on D0. Combining PCT and presence of fever may help to be more specific. Indeed, only PBI episodes presented both signs, whereas there was no PBI episode with the absence of both signs (Fig. 3). On D2, after administration of insulin and correction of glycemia, the near normalization of PCT and WBC in episodes without PBI may be explained by the correction of this inflammatory state, allowing to distinguish two different patterns: episodes with and without PBI. In the former group, episodes of fever occur and high levels of PCT, WBC, neutrophil count and NLCR still persist on D2. In the latter, there is a decrease, if not a normalization, of all the aforementioned markers following the correction of glycemia. Thus, on D0, infection status could be based on PCT level and presence of fever regardless of WBC or hypothermia occurrence. On D2, after normalization of glycemia, usual markers recover their discriminating potential and can enable the reassessment of antibiotic prescriptions if started on D0.

This study has some limitations. First, it was a monocentric retrospective analysis which limits the generalizability of the results. However, PCT and WBC measurement were systematically assessed for every patient admitted for DKA in our center. Second, the sample size is small with a significant number of excluded patients (32 out of 134) for different issues (as reported in Fig. 1) that may have may induced bias. Third, we only studied episodes of proven bacterial infection defined by a positive bacterial culture. This definition of bacterial infection may have induced bias, since cultures may not be realized or false-negative. Indeed, 31% of non-PBI episodes received antibiotics. Although this may point to overtreatment for some cases, it may also point to a non-proven, but still present bacterial infection. Thus, some episodes classified as without PBI may not be truly free from bacterial infection. Fourth, C-reactive protein was not available to almost all of the patients in our cohort. C-reactive protein was already reported to be associated with infection in patients with severe DKA [10]. However, as we did with PCT, some authors reported an elevation of C-reactive protein in patients free of infection during DKA episodes [39]. In our institution, C-reactive protein is not measured due to its contradictory result for critical care patients [40]. Even if C-reactive protein would have been informative, we were not able to include it in our study. Finally, temperature was not measured centrally. Central temperature device (urinary catheter, esophageal temperature probe or Swan–Ganz catheter) are not used in the context of DKA due to their invasive nature. In our institution, when central temperature is not available, axillary temperature is measured. The mean difference between axillary temperature and pulmonary artery temperature is known to be 0.5 °C (− 0.4 to 1.2 °C) [41]. In order to estimate core temperature, we choose to add a correction factor of 0.5 °C to the axillary measured temperature [20]. Despite these limitations, our study was a “real-life” observation of the population and was consistent with previous studies regarding triggering factors [2], survival rate [4] and increased length of stay in infected patients [9]. Prospective studies will be needed to confirm the interest and the diagnostic thresholds of these markers. Thus, a prospective clinical randomized control trial incorporating a decision rule based on procalcitonin and fever to guide the prescription of antibiotics during DKA could be conducted to confirm the clinical added value of such attitude.

At admission, our study showed that WBC, neutrophils count and hypothermia should not be taken into account in the diagnosis process of infection in diabetes ketoacidosis patients admitted in intensive care unit. Only fever and PCT (with a higher threshold than usual: 1.44 ng/mL) may help distinguish patients with and without PBI. By combining those two markers reduction of antibiotic misuse may be possible.