The fulminant index: A method of rapidly differentiating fulminant type 1 diabetes from diabetic ketoacidosis

Fulminant type 1 diabetes (FT1D) could present diabetes ketoacidosis (DKA) at early onset. It is crucial to identify FT1D from DKA manifestations in time at clinical practice. This study was aimed at investigating whether the fulminant index (FI), encompassing plasma glucose (PG) to glycated haemoglobin (HbA1c) ratio (PG/HbA1c), serum potassium ion (K+) to HbA1c ratio (K+/HbA1c) and serum sodium ion (Na+) multiplied by HbA1c (Na+*HbA1c), is a feasible indicator for early FT1D diagnosis.

muscle enzymes and even rhabdomyolysis, 7 leukaemia-like reaction, 8 sudden death, or cardiac arrest. 9 In that way, it could be necessary for clinicians to recognise FT1D early when handling DKA cases to identify highly risky cases and conduct the timely intervention. Besides, a fraction of FT1D cases caused by medication 10 or pregnancy 11 also requires early recognition to receive appropriate management. However, to our current knowledge, it remains a significant problem differentiating FT1D patients from general DKA patients in the emergency room. Current FT1D diagnosis mainly depends on Cpeptide monitoring. However, in the clinic, assessment of C-peptide could be infeasible under many circumstances, such as DKA emerging or patients' unconsciousness. In that way, we aimed to seek a better diagnostic pipeline for FT1D independent of C-peptide assessment, to achieve prompt differentiation and early management of potential FT1D patients who showed up in the emergency room. In recent years, an increasing number of FT1D cases have been reported in China, which has expanded our knowledge about the disease. 3,12,13 Compared to autoimmune T1DM, FT1D patients showed lower levels of serum sodium ion (Na + ) and glycated haemoglobin (HbA1c) and significantly higher levels of serum potassium ion (K + ) and plasma glucose (PG) 3,5 at early onset. Liu et al. 14 proposed a cutoff value of 4.2 for fulminant index (FI; PG/HbA1c), yielding 94% sensitivity and 98% specificity in differentiating FT1D from DKA. However, this study did not limit HbA1c in the DKA group. If HbA1c exceeded 8.7%, we could rule out FT1D without the PG/HbA1c in this study. Therefore, our current study established a limit for the HbA1c level in the DKA group and compared it with the FT1D group to test whether the PG/HbA1c ratio was effective and to explore whether there were better indicators for diagnosis.
Hyponatraemia and hyperkalaemia are commonly found in the DKA, and serum electrolyte disorders are more extensive in FT1D patients. As universally agreed in physiology, there is a negative correlation between Na + levels and PG and, conversely, a positive correlation between K + levels and PG. The underlying pathophysiology mechanisms may include the movement of electrolytes between intra-and extracellular spaces, impaired insulin action, as well as hyperosmolality. 15,16 Insulin could activate Na + /K + -ATPase. [17][18][19] The activity of Na + /K + -ATPase could be attenuated in insulindependent diabetic patients whose insulin secretion is impaired. In FT1D, hyponatraemia and hyperkalaemia could arise as a consequence of the remarkedly increase of plasma glucose and devastation of insulin-producing capacity. 5,12 In this study, we investigated multi-dimensional characteristics of 40 FT1D patients enroled since 2003. And we proposed a set of the quantitative diagnostic tool named FI, which was calculated based on PG, K + level, Na + level and HbA1c level. Since K + and PG were both higher in FT1D, we used the ratio of these two parameters with HbA1c to calculate FI. Na + and HbA1c were both relatively lower in FT1D, so we use FI (Na + *HbA1c) to indicate such changes in FT1D. We also calculated the estimated FI cutting-off points and verified its efficiency in differentiating FT1D from non-FT1D DKA patients through receiver operating characteristic (ROC) curve analysis.

| Patients inclusion and data collection
Our study included FT1D patients and non-FT1D DKA patients. We gathered the data of clinical characteristics of FT1D from literatures obtained in online database including CNKI database, Wanfang medical database, and PubMed database. The keywords used to search were 'fulminant type 1 diabetes mellitus' and 'Chinese'. A total of 206 articles and 654 cases were retrieved. Each patient should include a complete original record of the first visit. Patients with incomplete data were excluded, and the patients who were not the first treated were also excluded. In addition, FT1D patients should have to be matched with the age and gender of the control group. We retrieved the following information of each patient: demographic information including the gender and age, and clinical features or indexes including date of onset of hyperglycaemic symptoms. The following laboratory data were determined at the initial random measurement on admission: PG, electrolytes, urinary ketone bodies, blood gas analysis. Fasting blood samples were assessed within 24 h of admission for HbA1c levels. Serum C-peptide, β-cell autoantibodies, such as glutamic acid decarboxylase (GAD) antibodies, islet-associated antigen 2 (IA-2) antibodies, and insulin cell antibodies (ICAs), and diabetic complications were also recorded.
Ketosis was determined by urinary ketone bodies ≥2+. And β-cell autoantibodies were measured at the onset of disease. PG levels were tested using the glucose oxidase method. Electrolytes and blood gas analysis tested using the automatic biochemical analyser. In addition, HbA1c levels were tested using high-performance liquid chromatography through an automatic biochemical analysis system. In 23 cases from the literatures, HbA1c was tested using different methods including high-performance liquid chromatography (19/23) or immunochromatography (4/23).

| Study size calculation
To calculate the required sample size, a test for 1 ROC curve was applied to the data from our pilot study including 10 FT1D patients and 10 Non-FT1D DKA patients using PASS 11.0 software. Results suggested that 33 samples for each group were enough.

| Statistical analysis
All statistical analysis was performed using SPSS 16.0 software. Unpaired Student's t-test was used to analyse deviations of parameters between groups. The optimal FI cut-off values were calculated by ROC analysis using the Youden's index (sensitivity + specificity − 1).
Continuous variables were converted into binary variables based on the cut-off values. Logistic regression analysis was performed using these binary variables. We applied the area under the curve (AUC) to measure the diagnostic strength of FI. Sensitivity and specificity were assessed for the efficacy of FI in differentiating FT1D from non-FT1D patients. We also estimated the value of the combination of different FIs, which are determined by the Youden's index. For all computational analyses, p < 0.05 was considered statistically significant.
Continuous variables fitting normal distribution were described in the form of means � standard deviations (SDs).

| Clinical features of FT1D and non-FT1D DKA patients
The clinical features of FT1D and non-FT1D DKA were shown in 132.40 � 9.08 mmol/L; p = 0.001) were significantly lower in FT1D patients than that in non-FT1D DKA patients.
Between the groups, FI (PG/HbA1c) was significantly higher in    (Figures 1 and 2). Table 3, we tried to explore the value of the combination of different FIs in the diagnosis of FT1D. The value of the combination diagnosis is not entirely higher than that of the single FI.

| DISCUSSION
In clinical practice, DKA is one of the most encountered lethal complications. Nevertheless, for many reasons, clinicians may not be able to quickly and efficiently acquire the patients' previous diabetic QIU ET AL.

F I G U R E 1
Receiver operating characteristic curves for the fulminant index (PG/HbA1c, K + /HbA1c and Na*HbA1c) in the differential diagnosis between fulminant type 1 diabetes mellitus (FT1D) and non-FT1D we found FT1D patients showed lower levels of Na + and HbA1c and significantly higher levels of K + and PG than that in non-FT1D DKA patients, and FT1D patients showed lower levels of FI (Na + *HbA1c) and significantly higher levels of FI (K + /HbA1c) and FI (PG/HbA1c) than that in non-FT1D DKA patients.
As shown in our results, the AUC of these three FIs (PG/HbA1c, K + / HbA1c and Na + *HbA1c) is 0.830, 0.918 and 0.836, respectively. Youden's index of the FI (K + /HbA1c) is higher than that of both other FIs (PG/HbA1c and Na + *HbA1c). We can also construct other FIs based on the differences in the data for the two diseases, but they are too complex and no better than these. Among the FIs, we calculated, the K + /HbA1c ratio presented the best predicting efficiency, with superior AUC (0.918), and Youden's index (74.2%), which might be due to the physiology characteristic of serum K + . Both the endogenous insulin and the exogenous insulin are capable of affecting serum electrolyte levels, especially K + . Thus, DKA patients with insulin-dependent diabetes may not necessarily generate serum K + disorder if received insulin therapy.
Thus, to evaluate the patient metabolism environment accurately, we suggest assessing the FI (K + /HbA1c) before insulin intervention. For FT1D patients, those who have been treated with insulin may experience alleviation of hyperkalaemia, but the PG or serum Na + may not respond as well if the patient was severely dehydrated. So we tried to diagnose FT1D by using two or three of the FIs that meet FT1D. As can be seen from Youden's index in Table 3, FI (K + /HbA1c) was still the best predictor of FT1D. The combination of FIs (PG/HbA1c or K + /HbA1c) is the best diagnostic combination. So when FI (K + /HbA1c) fails to diagnose FT1D, we recommend using FI (PG/HbA1c).
In recent findings on the diagnosis of FT1D, GA, 21

| CONCLUSIONS
Our study suggested that the FI could work as a potential indicating index for identifying FT1D from general DKA patients. And

CONFLICT OF INTEREST
The authors have declared that no competing interest exists.

ETHICS STATEMENT
All procedures performed in studies involving human participants

DATA AVAILABILITY STATEMENT
The datasets generated and analysed during the current study are available from the corresponding author upon reasonable request.