Diabetic Ketoacidosis (DKA) is considered one of the most severe acute consequences of Diabetes Mellitus and is defined as the presence of hyperglycemia or existing diabetes mellitus, ketonemia, and metabolic acidosis. The latest consensus report from 2024 further clarified the current guidelines for diagnosis and treatment of DKA and highlighted the rising awareness of DKA in both type 1 and type 2 diabetes, such as those associated with infection, treatment interruption, acute illness, and sodium-glucose cotransporter-2 inhibitor use [1]. While standardized protocols for fluids, insulin and electrolytes have led to better outcomes of DKA, it remains a significant burden in the emergency department, inpatients and critical care units [1,2].

Clinically relevant early risk stratification because patients with the same glucose concentration can have a very different acid-base disturbance, inflammatory activation, renal dysfunction, hemodynamic compromise and clinical course. In recent years, the metabolic severity, organ dysfunction, presence of comorbidities and concurrent acute illness have been found to be factors associated with death or prolonged hospitalization in DKA [11-14]. A simple biomarker strategy in addition to the traditional laboratory markers (pH, bicarbonate, anion gap, ketone, mental status and renal function) might therefore be helpful.

The systemic immune-inflammation index (SII) is the product of three common complete blood count parameters (platelet count, neutrophil count and lymphocyte count), divided by each other. Biologic plausibility exists for SII as a marker of DKA severity, as neutrophilia, relative lymphopenia, cytokine activation, oxidative stress, and endothelial/platelet activation are found in DKA. In a cohort of patients with type 1 diabetes in 2024, SII was found to be higher as the severity of DKA increased, and better than simpler inflammatory ratios as a marker for severe DKA [3].

The link between inflammatory indices and severity of DKA, intensive-care course, and hospital stay has been established in subsequent studies [4,5]. The overall prognostic value of SII in diabetes-related complications and mortality outcomes was also supported by a 2025 systematic review and meta-analysis [6].

Serum lactate is a measure of a different—and complementary—physiologic area. The elevation of lactate in DKA may be due to tissue hypoperfusion, catecholamine-induced increased glycolysis, changes in redox state, decreased lactate clearance, or adaptive metabolism of the substrates. Previous research indicated that hyperlactatemia is frequent but not always associated with death or significant morbidity [7,8]. Recent research indicates that lactate kinetics or lactate in combination with another physiological parameter may have greater prognostic value. Impaired lactate clearance dynamics were associated with mortality in severe DKA in a 2026 multicenter prospective study [9] and the lactate to albumin ratio was found to be prognostic in a 2026 MIMIC-IV study [10].

The two markers may not be redundant, as SII gives information on the activation of inflammatory and thrombotic mechanisms and lactate gives information on metabolic/perfusion stress. Both are readily available from the tests that are often obtained during DKA evaluation. Therefore, the individual and combined prognostic value of admission SII and lactate was evaluated in this study for a clinically meaningful composite adverse outcome in adults with DKA.

2. Research Questions, Objectives, and Hypotheses

2.1 Research questions

• Do admission SII and serum lactate individually discriminate between adults with and without the primary composite adverse clinical outcome?
• Does a combined SII-lactate model show significantly better discrimination than either biomarker alone?
• Do SII and lactate remain independently associated with the primary composite outcome after adjustment for demographic and clinical covariates?
• Are SII and lactate associated with ICU/HDU admission, delayed DKA resolution, hospital length of stay, AKI, 30-day readmission, and a harder adverse-outcome composite?
• Which exploratory thresholds maximize the Youden index for SII, lactate, and the combined model, and how do observed adverse-outcome rates differ across threshold-defined groups?

2.2 Primary objective

To assess the ability of the combined model of admission SII and serum lactate to better discriminate the primary composite adverse clinical outcome than admission SII or serum lactate alone.

2.3 Hypotheses

• H1: The combined SII-lactate model will have a significantly larger AUC than either SII or lactate alone.
• H2: Higher SII will be independently associated with increased odds of the primary composite outcome after multivariable adjustment.
• H3: Higher serum lactate will be independently associated with increased odds of the primary composite outcome after multivariable adjustment.
• H4: Higher SII and lactate will be positively associated with slower DKA resolution and longer hospital stay and will retain prognostic signal for a harder clinical composite.
• H0: The combined model will not improve discrimination over either individual biomarker and neither biomarker will remain independently associated with the primary outcome after adjustment.

3.1 Study design and reporting approach This study was designed as a multicenter observational cohort study of adults with DKA. The Strengthening of Reporting of Observational Studies in Epidemiology (STROBE) statement [17] was followed for reporting. The work also included an evaluation of a multivariable prognostic model and therefore the reporting of the model specification, discrimination, calibration and need for independent validation were adhered to the TRIPOD+AI principles [16]. 3.2 Study population and setting Analytic sample consisted of 410 adults (age 18 years or older) with DKA. The cases were spread over nine site categories within Karachi, Lahore, Rawalpindi, Islamabad, Faisalabad, Multan, Peshawar, Hyderabad and Quetta. The eligibility and severity of DKA were determined using the current international consensus definition of diabetes/hyperglycemia, ketonemia, and metabolic acidosis [1,15]. 3.3 Clinical and laboratory variables The variables included in this study were age, sex, diabetes type and duration, newly diagnosed diabetes, infection as the precipitating factor, Charlson Comorbidity Index (CCI), Glasgow Coma Scale (GCS), glucose, beta-hydroxybutyrate (BHB), pH, bicarbonate, anion gap, creatinine, estimated glomerular filtration rate (eGFR), potassium, platelet count, neutrophil count, lymphocyte count, lactate, DKA resolution time, and hospital length of stay. DKA severity was classified as mild, moderate or severe, and based on the present concepts of severity of DKA [1,15]. 3.4 Biomarker definitions To results with current SII literature in DKA and diabetes [3-6] and confirm the validity of the SII measurement, we calculated SII as platelet count x neutrophil count / lymphocyte count. SII was calculated using the model based upon a 1000-unit increase in order to enhance the interpretability of odds ratios. Prior and recent studies of DKA (7-10) informed the interpretation of admission lactate as a metabolic/perfusion marker. The two-marker combined logistic equation was: logit(p) = -4.527 + 0.302 x (SII/1000) + 0.880 x lactate The model-generated predicted probability was used for ROC analysis. Discrimination and calibration measures were used to summarize apparent model performance as per modern reporting guidelines for prediction models [16]. 3.5 Outcome definitions The primary outcome was a composite adverse clinical event that included in-hospital death, vasopressor/inotropic use, prolonged DKA resolution, mechanical ventilation or ICU/HDU admission during the index admission. The individual components, DKA resolution time, hospital length of stay, acute kidney injury (AKI), and 30-day diabetes-related readmission were considered secondary outcomes. A sensitivity outcome did not include prolonged DKA resolution, but did include admission to an ICU/HDU, mechanical ventilation, vasopressor/inotropic support, or death during hospitalization. Clinically meaningful severity and organ-support outcomes are similar to recently published DKA outcome literature [1,12-15]. 3.6 Missing data The 410 cases with complete data for the primary outcome status, SII, lactate, and the variables in the main adjusted model were analysed. For 372 cases, BHB was available. Thirty-day readmission data was available for 389 cases, with no cases missing. Valid observations were analysed for each procedure; not imputing missing values. 3.7 Statistical analysis Continuous variables were expressed as mean +/– standard deviation and categorical variables as frequency/percentage. Normality diagnostics and graphical inspection were used to explore distributional characteristics. Independent-samples testing along with distribution-robust sensitivity comparisons (Mann-Whitney U tests) were used for continuous variables between outcome groups. The categorical variables were compared using the Pearson chi-square or Fisher exact tests, whichever was appropriate. Primary analysis p values < 0.05 on both sides were deemed statistically significant. The extent of discrimination of SII, lactate and the combined predicted probability was quantified using ROC analysis with AUCs and 95% confidence intervals. Correlated AUCs were compared to the non-parametric DeLong method [18]. The thresholds used for exploratory were determined by the Youden index. Results were presented as odds ratios (OR) with 95% confidence intervals (CIs) using logistic regression. SII/1000, lactate, age, sex, diabetes type, CCI, infection precipitant, admission pH and GCS were included in the main adjusted model. Observed versus mean predicted risk across deciles with the Hosmer-Lemeshow test were used to assess calibration, as per guidance from TRIPOD+AI [16] with the focus on discrimination and calibration. Secondary analyses included tests of ICU/HDU admission, AKI, 30-day readmission, and Spearman correlations with time to DKA resolution and hospital LOS, and the harder-composite sensitivity outcome. If the secondary analyses were not interpretable, they were considered exploratory.

4.1 Cohort characteristics

The cohort included 410 adults with a mean age of 41.80 +/- 15.08 years; 231 (56.3%) were male. Type 2 diabetes accounted for 221 (53.9%) cases, type 1 diabetes for 170 (41.5%), and other/undetermined diabetes for 19 (4.6%). Forty-five (11.0%) were newly diagnosed and 121 (29.5%) had infection as a precipitating factor. DKA severity was mild in 153 (37.3%), moderate in 175 (42.7%), and severe in 82 (20.0%).

Table 1. Baseline demographic, clinical, laboratory, and site characteristics (N=410)

Values are mean +/- SD or n (%), unless otherwise indicated. BHB was available for 372 cases. CCI, Charlson Comorbidity Index; GCS, Glasgow Coma Scale; SII, systemic immune-inflammation index.

4.2 Primary and secondary outcome incidence

The most common composite adverse event was in 86/410 (21.0%) patients. Prolonged DKA resolution was the most frequent component (72/410; 17.6%), followed by ICU/HDU admission (28/410; 6.8%), vasopressor/inotropic support (12/410; 2.9%), mechanical ventilation (4/410; 1.0%), and in-hospital death (1/410; 0.2%). AKI occurred in 52/410 (12.7%). Thirty-day readmission rate was 32/389, or 8.2% (where follow up data was available). The adverse composite was harder to find, and was present in 37/410 (9.0%).

Table 2. Clinical outcome incidence

Figure 1. Incidence of the primary composite, its components, AKI, 30-day readmission, and the harder adverse composite.

4.3 Continuous variables according to primary composite outcome

The primary composite outcome was used as the basis for identifying continuous variables.Continuous variables were identified based upon the primary composite outcome. The adverse-outcome group exhibited more distressed metabolic and inflammatory responses. The values of the following indices were lower: GCS, pH, bicarbonate, eGFR, and lymphocyte count; and higher: glucose, BHB, anion gap, creatinine, neutrophil count, SII, lactate, DKA resolution time, and hospital length of stay. The ages, CCI and potassium and platelet counts were not significantly different in Mann-Whitney analysis. The direction of changes in the SII and lactate differences is consistent with the results of recent studies on DKA biomarkers, which showed increased inflammatory activation and metabolic stress to be associated with increased severity or poorer clinical outcomes [3-5,9,10].

Table 3. Continuous variables by primary composite adverse outcome

Values are mean +/- SD. The p values shown are the distribution-robust Mann-Whitney tests; independent-samples analyses showed the same overall direction of the major findings.

4.4 Categorical associations with the primary composite outcome

The rate of adverse outcomes was significantly higher in moderate DKA compared to mild DKA (9.8 % to 22.3 %; Pearson chi-square=27.816, p<0.001) and severe DKA compared to moderate DKA (22.3 % to 39.0 %; Pearson chi-square=11.25, p=0.001). Higher outcome rate was found for those who had infection as a precipitating factor (33.9% vs 15.6%; chi-square = 17.257, p<0.001). Baseline eGFR <60 mL/min/1.73 m2 was also associated with a higher adverse-outcome rate (29.0% vs 18.6%; p=0.030). There was no significant association between sex and type of diabetes. These patterns are in keeping with current data highlighting the severity of DKA, precipitating illness, renal dysfunction and the presence of comorbidity as factors affecting clinical course [1,12-15].

Table 4. Categorical characteristics and primary adverse-outcome rates

Figure 2. Primary composite adverse-outcome rates across DKA severity, infection-precipitant, and renal-function strata.

4.5 ROC analysis and exploratory thresholds

SII showed an AUC of 0.733 (95% CI 0.676-0.791; p<0.001), and lactate showed an AUC of 0.747 (95% CI 0.692-0.802; p<0.001). There were no significant differences between their individual AUCs (p=0.670). The combined SII-lactate probability increased the AUC to 0.787 (95% CI 0.732-0.842; p<0.001). The improvement was significant versus SII alone (AUC difference 0.054, 95% CI 0.013-0.094; p=0.009) and versus lactate alone (difference 0.040, 95% CI 0.010-0.070; p=0.010). The rationale behind this incremental discrimination is biological: inflammatory and metabolic/perfusion domains are distinct, and are captured by SII and lactate, respectively. The Youden-optimal exploratory thresholds were a combined predicted probability of 0.2821 (sensitivity 60.5%, specificity 84.9%), lactate 2.25 mmol/L (sensitivity 82.6%, specificity 57.4%), and SII 2324.18 (sensitivity 72.1%, specificity 62.0%).

Table 5. ROC performance for the primary composite outcome

Thresholds are internally derived and should not be interpreted as validated clinical decision cutoffs.

Figure 3. Receiver operating characteristic curves for SII, lactate, and the combined SII-lactate model.

Figure 4. Comparison of AUC values for SII, lactate, and the combined model.

4.6 Threshold-based classification and calibration

At the internally derived SII threshold, 62 of 185 test-positive cases experienced the primary adverse outcome (33.5%) compared with 24 of 225 test-negative cases (10.7%). Of 209 test-positive cases, 71 (34.0%) had the outcome, while 15 of 201 test-negative cases had the outcome (7.5%). These contrasts do not provide an estimate of sensitivity and specificity and are not established clinical cut points but are exploratory [16].

Table 6. Primary adverse-outcome rates by exploratory biomarker threshold status

Threshold status uses the internally selected cut points corresponding to the ROC analyses. These values require external validation before clinical use [16].

Figure 5. Primary adverse-outcome rates in test-negative and test-positive groups at the exploratory SII and lactate thresholds.

Calibration of the fully adjusted model was also analysed at the 10 deciles of predicted risk. In general, mean event rates increased as mean predicted risk increased, ranging from 2.0% observed to 3.0% predicted in the lowest-risk decile, to 68.0% observed to 64.9% predicted in the highest-risk decile. The seventh decile exhibited the greatest difference that could be seen (15.0% observed - 21.9% predicted). The decile pattern did not indicate gross miscalibration, in conjunction with a Hosmer-Lemeshow p-value of 0.944, and formal external calibration is required [16].

Table 7. Observed and predicted primary adverse-outcome risk by decile of the adjusted model

Each decile contained 41 cases. Decile-based calibration summaries describe apparent performance in this cohort [16].

Figure 6. Observed and mean predicted primary adverse-outcome risk across deciles of the fully adjusted model.

4.7 Two-marker and multivariable logistic regression

In the two-marker model, each 1000-unit increase in SII was associated with higher odds of the primary composite outcome (OR 1.353, 95% CI 1.186-1.543; p<0.001), and each 1 mmol/L increase in lactate was associated with higher odds (OR 2.412, 95% CI 1.747-3.329; p<0.001). The model likelihood-ratio test was significant (chi-square=76.728, 2 df, p<0.001), the Nagelkerke R2 was 0.266 and the Hosmer-Lemeshow test was not significant (p=0.899). Overall classification accuracy was 82.2% at a probability cut point of 0.50, with 96.3% correct classification of non-events and 29.1% correct classification of events. Both markers provide independent prediction, which is in line with the literature on inflammatory and lactate-based markers for prognosis in DKA [3-5,9,10]. In the adjusted model, SII remained independently associated with the composite outcome (aOR 1.327 per 1000 units, 95% CI 1.147-1.535; p<0.001), as did lactate (aOR 2.442 per mmol/L, 95% CI 1.666-3.580; p<0.001). Age, sex, diabetes type, CCI, infection precipitant, pH and GCS were not found to be independent significant factors. The adjusted model had a likelihood-ratio chi-square of 81.549 (10 df; p<0.001), Nagelkerke R2 of 0.281, Hosmer-Lemeshow p=0.944, and AUC 0.793 (95% CI 0.740-0.847). As per TRIPOD+AI, these values are not used as a final clinical score but rather as apparent performance, which needs to be verified independently.

Table 8. Multivariable logistic regression for the primary composite adverse outcome

4.8 Secondary outcome models

For ICU/HDU admission, SII was significant in the two-marker model (OR 1.297 per 1000 units, 95% CI 1.117-1.506; p=0.001), whereas lactate was borderline (OR 1.532, 95% CI 0.990-2.370; p=0.055). For AKI, neither SII (aOR 0.960, 95% CI 0.793-1.161; p=0.672) nor lactate (aOR 1.471, 95% CI 0.971-2.227; p=0.068) was significant after adjustment. Age was associated with AKI (aOR 1.046 per year, 95% CI 1.012-1.082; p=0.008); diabetes type had an overall p=0.028, with type 2 diabetes showing lower odds than type 1 diabetes (aOR 0.330, 95% CI 0.136-0.800; p=0.014).

The mixed secondary-outcome pattern is logical for the heterogeneity of DKA prognostic studies; in these studies, inflammatory and metabolic markers will differ based on endpoint definition and case mix [4,5,9,12,13]. Overall, the two-marker readmission model was statistically significant (likelihood-ratio chi-square=6.080, 2 df; p=0.048) with a low explanatory performance (Nagelkerke R2=0.036) in the 389 cases with a 30-day follow-up. SII showed an inverse association with readmission (OR 0.713, 95% CI 0.518-0.982; p=0.038), whereas lactate was not significant (OR 1.352, 95% CI 0.882-2.074; p=0.167). Due to the nature of the analysis (secondary analysis, weak model, and missing follow up), the result should be interpreted with caution. Finally, too few patients died (n=1) or received vasopressor/inotropic support (n=12) for stable, independent multivariable modeling, and therefore these are only reported descriptively. Even this interendpoint variation underscores the importance of not extrapolating the relationship of one biomarker across all DKA outcomes [12,13,16].

Table 9. Biomarker associations with selected secondary binary outcomes

AKI estimates are from the adjusted model. ICU/HDU and readmission estimates are from two-marker models.

Figure 7. Model-based odds ratios for SII and lactate across the primary and selected secondary outcomes. Model specifications are identified in the category labels.

4.9 Correlations with DKA resolution and hospital length of stay

There was a moderate correlation between SII and lactate (Spearman rho=0.432, p<0.001). SII correlated with DKA resolution time (rho=0.528, p<0.001) and hospital length of stay (rho=0.387, p<0.001). Lactate also was correlated with time for resolution of DKA (rho=0.556, p<0.001) and hospital length of stay (rho=0.454, p<0.001). Duration of stay and time to resolve DKA was significantly correlated (rho=0.625, p<0.001). The positive SII-duration association is similar to recent DKA work relating inflammatory indices to hospitalization [4,5] and the lactate association is directionally similar to more recent work focusing on lactate as a risk indicator [9,10].

Table 10. Spearman correlation analysis

Figure 8. Spearman correlations of SII and lactate with DKA resolution time and hospital length of stay.

4.10 Hard-composite sensitivity analysis

37% of 410 cases (9.0%) were the harder composite. SII had an AUC of 0.747 (95% CI 0.669-0.825; p<0.001) and lactate an AUC of 0.721 (95% CI 0.634-0.807; p<0.001). In a two-marker logistic model, SII remained associated with the harder outcome (OR 1.344 per 1000 units, 95% CI 1.167-1.548; p<0.001) and lactate also remained associated (OR 1.977 per mmol/L, 95% CI 1.315-2.973; p=0.001). The likelihood-ratio chi-square was 37.364 (2 df; p<0.001), Nagelkerke R2 was 0.192, and Hosmer-Lemeshow p=0.521. The persistence of the signal following prolonged resolution is in line with previously published data that suggest that SII and lactate-related measures are associated with more severe clinical outcomes of DKA [3,9,10].

Table 11. Sensitivity analysis for the harder adverse-outcome composite

4.11 Hypothesis evaluation

Table 12. Evaluation of prespecified hypotheses

5.1 Principal findings The prognostic signal of SII and lactate were consistent across centers in adult DKA. Patients with the primary composite outcome showed higher SII and lactate levels and more severe acidemia (lower pH and bicarbonate), elevated anion gap and creatinine levels, and significantly prolonged DKA resolution and hospitalization. The combination of the individual biomarkers resulted in a significant improvement in AUC, with moderate discrimination for the individual biomarkers. Importantly, the association of SII and lactate with the primary composite was independent of age, sex, diabetes type, comorbidity burden, infection precipitant, pH, and GCS. These results are in line with the recent evidence related to severity of SII-related DKA and risk stratification by lactate level [3-6,9,10]. 5.2 SII and the inflammatory component of DKA severity The results of the SII were similar to those found in DKA - the inflammatory phenotype. Aon et al. identified a rising trend of SII as the severity of DKA, and found that it had a better discriminatory ability than the neutrophil-to-lymphocyte ratio and the platelet-to-lymphocyte ratio [3]. The inflammatory markers were then found to be associated with the severity of DKA and clinical outcomes in adult patients by Aslan Sirakaya et al. [4] and with hospital stay by Karaduru Avci and Sargin [5]. The present findings confirm this trend of associations of severity and length of stay with a broader adverse clinical composite. This larger 2025 meta-analysis of SII in diabetes also reinforces the notion that systemic inflammatory burden is related to diabetic complications and diabetes deaths [6]. 5.3 Lactate and metabolic/perfusion stress actate has a complex meaning in DKA. Cox et al. and Bhat et al. demonstrated that lactate elevation is frequent and could be associated with change in metabolism as well as hypoperfusion, which could account for the lack of consistent association with hard outcomes with a single lactate value [7,8]. More recent studies have bolstered the prognostic value of lactate dynamics or composite lactate values. Sebastian-Valles et al. showed that lactate level behavior in prehospital setting was a predictor of mortality in severe DKA [9] and Zhang et al. showed that the lactate to albumin ratio was associated with 28-day mortality among critically ill patients with DKA [10]. The present findings support the concept that lactate provides valuable prognostic information when viewed in a physiologic perspective and not as a marker for shock. 5.4 Why the combined model performed better It is plausible that the SII and lactate were additive and that the incremental AUC increment is an example of complementarity across the pathways. Inflammatory balance of leukocytes and platelets is mainly captured by SII, whereas metabolic and perfusion stress is captured by lactate. There is partial correlation between them (rho=0.432) and this means that there is some overlap and not repetition. The retained independent coefficient(s) in the primary models, and the combined AUC was higher than either single-marker AUC(s). This reflects the overall predictive reasoning of some of the more common laboratory ratios used in predicting outcomes of combined laboratory values in acute DKA and critical illness, such as BUN to albumin and lactate to albumin [10,11]. 5.5 Relationship with DKA severity, infection, and renal dysfunction The incidence of adverse outcomes increased in a graded manner across all DKA severity groups, was more than double when there was an infection as a precipitating event, and was higher in the presence of baseline renal impairment. These univariable patterns match clinical reality and the consensus and literature on contemporary DKA, which have highlighted the role of precipitating illness, severity of acid-base disturbances, organ dysfunction and comorbidity in determining clinical trajectory [1,12-15]. It should not be interpreted that these conventional covariates are unimportant clinically because of the loss of its independent significance in the primary adjusted model; this may be because of shared information, limited events and strong association of the biomarker pair with the composite endpoint. 5.6 Secondary outcomes and clinical implications SII was associated with ICU/HDU admission, and both biomarkers were associated with longer DKA resolution and LOS. Both SII and lactate were maintained, and showed significant associations, despite the use of the harder-composite sensitivity analysis, indicating that the primary signal was not only a prolonged resolution of DKA. The analyses of AKI and readmission, however, were not as consistent. The inverse SII relationship with 30-day readmission was obtained in secondary analysis of a model with low power of explanation and should be interpreted as a hypothesis rather than as definite conclusions for clinical practice. The most likely use for an SII-lactate model in DKA is if it is reproduced, as an adjunct to existing bedside assessment, not instead of standard markers of DKA severity. It may help to facilitate discussions on monitoring intensity or escalation during early monitoring because the two inputs are readily available from routine testing. Yet, independent testing, validation, calibration, testing in subgroups and decision thresholds/clinical usefulness must be considered, as stated by TRIPOD+AI [16]. 5.7 Strengths and limitations The advantages of this study are the simultaneous measurement of two low-cost biomarkers that measure different biological domains, a well-defined composite endpoint, paired ROC comparison, multivariable adjustment, calibration reporting, reporting of secondary endpoints, and the absence of a prolonged resolution sensitivity composite. The results of the group comparisons, ROC analysis, regression coefficients, correlations, and hard-composite analysis increase the internal consistency of the results. There are a few restrictions to be taken into consideration. The development and evaluation of the prediction model was done in the same cohort, which may lead to an optimizing performance and needs to be validated externally [16]. Clustering at site level was not modeled; serial lactate kinetics were not available; BHB was not available for 38 cases; and follow-up after readmission was not complete in 21 cases. The event numbers of mechanical ventilation and in-hospital mortality were very small, which prevented stable regression of the endpoint specific. The harder composite had just 37 events, restricting the complexity of the models. Treatment variables and response over time were not included. Future studies must incorporate internal resampling, external geographic validation, calibration slope and intercept, decision-curve analysis and prespecified subgroup interactions. 6. Summary of Findings • Twenty-one percent (410) of cases had the main composite adverse outcome. • The SII and lactate level and the degree of metabolic derangement were higher in patients with adverse outcomes. • Moderate discrimination was obtained for SII and lactate alone, and was enhanced by their combination to 0.787. • AUCs for combined SII and lactate was significantly greater than for SII alone and lactate alone, and threshold-defined positive groups had significantly higher observed adverse-outcome rates than negative groups. • Multivariable adjustment did not remove the independent association between SII and lactate with the primary composite; the adjusted model had AUC 0.793 and the observed and predicted risk was generally concordant within deciles. • Both biomarkers were associated with delayed DKA resolution and longer hospital stays and remained associated with a composite of adverse outcomes, even after adjustment. • Secondary AKI and readmission analyses were less consistent and need to be viewed with caution.

The variables of admission SII and serum lactate provided complementary information for predicting prognosis in adults with DKA. They worked synergistically to increase the ability to discriminate a clinically significant composite adverse outcome, and both were independently associated with outcome after multivariable adjustment. These results provide encouragement to further test an affordable strategy of early DKA risk stratification based on inflammation plus metabolic stress. The model and thresholds must be validated, calibrated, and clinically-utility assessed independently before they can be used in patient care [16].

Authors Role: