For the purpose of this retrospective study, the medical records of 210 calves aged between one and 5 months that had a diagnosis of diarrhea and were admitted to the Clinic for Ruminants with Ambulatory and Herd Health Services, LMU Munich, between May 2005 and August 2023 were reviewed. Calves were identified using the clinic’s hospital record database. Calves were selected for inclusion in this study if they were admitted for treatment of diarrhea, had a positive test result for Eimeria spp. oocysts on a fecal examination (flotation technique), and if blood gas, electrolyte and clinical biochemical analyses were performed on admission. As in previous studies, diarrhea was defined as a fecal consistency that permitted feces to run through slightly opened fingers (10, 13). General exclusion criteria for calves were a positive test results for BVDV antigen and incomplete datasets regarding blood gas, electrolyte, and biochemical analyses.

The clinical information retrieved from the medical records included signalment (age, sex, breed), whether the calf was already weaned or not, respiratory and heart rate, rectal temperature in °C, presence or absence of injected scleral blood vessels, hyperemia of mucous membranes, presence of bloody diarrhea, and presence of tenesmus. As in a previous study (13), the following parameters were categorized using 3-point scales on the initial examination at admission: posture (ability to stand, impaired ability to stand, sternal or lateral recumbency), behavior (bright and alert, depressed, apathetic to comatose), suckling reflex (strong, weak, absent), degree of enophthalmos (none, slight to moderate, severe), skin tent duration assessed on the upper eyelid (normal, slightly to moderately increased, markedly increased), and body condition (good to moderate, poor, cachectic). Also as in previous studies (13, 26), a diagnosis of systemic inflammatory response syndrome on hospital admission was made if two of the following criteria were fulfilled: presence of an abnormal leukocyte count (< 5 or > 12 × 10⁹ cells/L), abnormal rectal temperature (< 38.5°C or > 39.5°C), tachycardia (> 120 beats per minute), and tachypnea (> 36 breaths per minute). Information regarding therapeutic measures, presence of concurrent health problems during the first 48 h of hospitalization, outcome of therapy, post-mortem findings, and the stated reasons for euthanasia was also extracted.

Blood samples were collected from the jugular vein on admission to the hospital before any treatment was administered. Lithium-heparinized blood samples were anaerobically collected using a 2-mL polypropylene syringe and blood pH, pCO₂, and concentrations of plasma sodium (cNa⁺), chloride (cCl⁻), potassium (cK⁺), and ionized calcium (cCa²⁺) were determined using blood pH, gas, and electrolyte analyzers (Rapidlab 865 [2005–2012], Rapidpoint 405 [2012–2020], and Rapidpoint 500 [2020–2023], Siemens Healthcare Diagnostics Inc., Tarrytown). Blood pH and pCO₂ were corrected for rectal temperature using the same equation for all three analyzers.

Automatic analyzers were used for additional serum and plasma biochemical analyses (Hitachi 911 [from 2005–2012] and Cobas c311 [from 2012–2023] analyzers, Roche Diagnostics, Mannheim, Germany). Serum samples (plain tubes) were assayed for concentrations (c) of urea (urease), creatinine (picric acid), total protein (biuret), albumin (bromcresol green), and inorganic phosphorus (molybdenum). Blood samples containing lithium heparin and potassium fluoride as a glycostatic agent were analyzed for plasma concentrations of L-lactate (lactate oxidase), glucose (hexokinase) and in some single calves for D-lactate using a D-lactate dehydrogenase assay (27). In 16 calves, serum samples, that had been stored at −25°C up to 3 years, were additionally analyzed for D-lactate (using the same D-lactate dehydrogenase assay) and ß-hydroxybutyrate (hydroxybutyrate dehydrogenase) concentrations. Fecal samples were analyzed for Eimeria oocysts using the flotation technique. Between 2005 and 2012, EDTA blood samples were routinely tested for BVDV-antigen in all hospitalized calves (PCR technique), which was stopped thereafter due to major advances in the national BVDV control program.

Application of the Henderson-Hasselbalch acid–base model included assessment of actual bicarbonate concentration (cHCO₃⁻) and base excess of blood (abbreviated as BE(B), also called in vitro base excess) which were automatically calculated by the blood gas analyzers based on previously reported algorithms (10, 13). Anion gap (AG) was calculated in order to quantify the unmeasured anion concentration in plasma based on the following equation (28):

The simplified quantitative physiochemical strong ion approach (29) was additionally applied as it provides a more mechanistic approach to acid–base balance and consequently a more comprehensive assessment of the acid–base status of calves than it is the case for the Henderson-Hasselbalch acid–base model (30, 31). This included calculation of the measured strong ion difference obtained from five strong ions (SID₅, mEq/L) using the measured value for [Ca²⁺] determined by ion-selective potentiometry and assigning a charge of −1 to L-lactate assuming 100% dissociation such that (31):

To be able to assess the relative importance of sodium and chloride imbalances on acid–base status, the strong ion difference calculated from the plasma concentrations of sodium, potassium and chloride (SID₃; mEq/L) was additionally obtained as follows (9):

An estimate of the concentration of non-volatile weak acids (A_tot; consisting of albumin, globulins, and dihydrogen phosphate) in mmol/L was estimated from serum concentrations of total protein as previously described (9):

The anion charge of non-volatile weak acids (A⁻) was calculated using the experimentally determined value for the negative logarithm of dissociation constant of plasma nonvolatile weak acids (pK_a = 7.08) and the following equation (9):

The unmeasured strong ion concentration was obtained by calculating the strong ion gap (SIG) in mEq/L, defined as the difference between the plasma concentration of unmeasured strong cations and unmeasured strong anions. In contrast to the traditional anion gap (Equation 1), this approach excludes the effect of the anion charge of non-volatile weak acids, whereby (9):

Calculated values for SIG were also corrected for the measured concentrations of L-lactate and Ca²⁺ to obtain the concentration of still unidentified strong ions (USI) in mEq/L for the calculation of the effective strong ion difference (SID_eff) such that (31):

Measured concentrations of plasma ionized calcium were corrected for changes in pH from 7.40 by use of the following formula (32):

Furthermore, a potassium to sodium ratio was calculated by dividing cK⁺ with cNa⁺ and multiplying values with 100 (33).

In order to detect disproportional changes of plasma chloride concentrations with regards to changes of plasma sodium concentrations, a corrected plasma chloride concentration (cCl_c⁻) was calculated (34, 35) using the following formula, where normal cNa⁺ is the midpoint (142 mmol/L) of the reference interval for plasma sodium concentration:

Positive outcome of treatment (PO; survival) was defined as discharge from the hospital, whereas a negative outcome (NO; non-survival) was defined as death or euthanasia during hospitalization.

Statistical analyses were performed using SPSS for windows (version 26.0, IBM Corp., Armonk, New York, United States), GraphPad Prism (version 10.1.1, GraphPad software) and the software package R (version 4.3.2., R Core Team). Normality of data was assessed based on results of a Shapiro–Wilk test and visual examinations of QQ plots. As most of the data were not normally distributed, non-parametric tests were used and data are presented as medians and interquartile ranges (Q₁/Q₃).

Spearman′s correlation coefficients (r_s) were calculated in order to assess potential associations between parameters. Stepwise forward variable selection for linear regression models were additionally constructed in order to predict venous blood pH and actual bicarbonate concentration by variables of clinical pathology and independent variables of the strong ion difference acid–base model (SID, A_tot, pCO₂). For the latter, regression models were calculated twice based on SID₃ and SIG as well as SID₅ and USI as representatives of the effective strong ion difference. The relative importance of the included variables was assessed by the order of entry into the model as well as by the change in the model R² value (ΔR²). In case of close associations between independent variables (r_s > 0.70 or < −0.70), only one variable was entered into the model in order to minimize potential effects of collinearity (10). Standardized residual plots of multivariable models were examined to confirm an approximately normal distribution of residuals. Multicollinearity of the final predictors was assessed by calculation of the variance inflation factor; multicollinearity is considered a concern if the variance inflation factor is higher than 10 (36).

Mann–Whitney U-tests were used for pair-wise comparison of continuous variables. Differences between different categories of posture/ability to stand were assessed using a Kruskal-Wallis test. Univariable associations between clinical and laboratory variables and the outcome of therapy was assessed using binary logistic regression analysis, which included calculation of odds ratios (OR) with 95% confidence intervals (95% CI). For multivariable modelling of prognostic factors, predictors with a p-value ≤0.2 were further processed via a novel “best subset”—automated model selection and multimodel inference approach. This approach, implemented in the “glmulti” statistical package (37), finds the best set of models amongst all possible models. Specifically, best subset approach evaluated 42,152 models resulting from various combinations of five predictors and an additional 1,450 models generated from all possible pairwise interactions between 20 predictors. The “best subset” approach ranked these models based on their Akaike Information Criterion (AIC) values, identifying the optimal model. Subsequently, we compared this model’s AIC with that of the “final” model obtained through a backwards-selection approach, aiming to contrast the two variable selection methodologies, and since the final model of the best subset has a significantly better AIC (AIC difference > 2) as compared to the final model resulting from the backwards selection, the best subset approach was opted for. The fit of the final multivariable regression model was evaluated by means of the Hosmer-Lemeshow goodness-of-fit test. The final multivariable models was also assessed for potential confounding effects (38) by the presence of concurrent health problems.

Data were additionally analyzed using classification tree analysis to identify clinically relevant cut-points as potential mortality predictors. The predictive ability of the models was assessed by calculating the area under the receiver-operating characteristic (ROC) curve and the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) at the optimal cut-point of predicted probabilities identified on the basis of the Youden index. p-values <0.05 were declared as statistically significant, while p-values between 0.05 and 0.1 were treated as tendentially significant for all analyses.

A total of 136 calves met the inclusion criteria for the study out of 210 calves aged between one and 5 months that had a record of diarrhea in the clinic’s hospital record database during the elected study period. Of those 136 calves, four calves were excluded due to a positive BVDV test result, 13 calves were excluded due to an incomplete dataset regarding laboratory variables, and one calf that was presented in an agonal state was excluded due to documentation issues. Therefore, a total of 118 calves was used for the analysis reported here. The study population consisted of German Fleckvieh (n = 112; 94.9%) and Braunvieh calves (n = 6; 5.1%). The median age (Q₁/Q₃) was 8.4 (6.6/11.0) weeks, and 23 calves (19.5%) were already weaned on admission to the hospital. The proportion of female and male calves was 65.3% (n = 77) and 34.7% (n = 41), respectively.

Most calves were presented in a critical condition. The majority of calves had bloody diarrhea (n = 95; 80.5%) and exhibited tenesmus (n = 66; 55.9%). A rectal prolapse was documented in 6 calves on admission. Clinical dehydration (increased duration of skin tenting or presence of enophthalmos) was documented in 97 calves (82.2%). Alterations of behavior and posture/ability to stand were documented in 72 calves (61%) and 50 calves (42.4%), respectively. At least one concurrent health problem was documented during the first 48 h of hospitalization in 31 calves (26.3%), which included pneumonia (n = 10), lice infestation (n = 6), clinical signs of acute abdomen (n = 4), left-sided displacement of the abomasum (n = 3), navel infections (n = 3), septic arthritis (n = 2), and miscellaneous problems (n = 4).

Median and interquartile ranges for selected acid–base and clinicopathologic variables and respective reference ranges are presented in Table 1. The most common clinicopathologic abnormalities included hyponatremia (n = 107; 90.7%), hypochloremia (n = 90; 76.3%), azotemia (n = 82; 69.5%), hyper-L-lactatemia (n = 73; 61.9%), and hyperglycemia (n = 84; 71.2%). Out of the 107 hyponatremic calves, 61 calves had plasma sodium concentration between 105 and 120 mmol/L, and 14 calves <105 mmol/L. Hyponatremia was accompanied by hypochloremia in 89 calves. As shown in Figure 1A, plasma sodium and chloride concentrations were strongly correlated (r_s = 0.90; p < 0.001). Sodium corrected plasma chloride concentrations (calculated using Equation 10) were below and above the reference range in 11 and 12 calves, respectively. Individual values for serum urea and creatinine concentrations are shown in Figure 1B.

Higher than normal serum inorganic phosphorus concentrations (>3.1 mmol/L) were present in 40 calves (33.9%). Hyperkalemia (plasma potassium >5.8 mmol/L) was present in 34 calves (28.8%), whereas hypokalemia (plasma potassium <3.9 mmol/L) was present in 17 calves (14.4%). Plasma potassium concentrations were most closely correlated to serum creatinine (r_s = 0.57; p < 0.001) and inorganic phosphorus (r_s = 0.56; p < 0.001) concentrations, respectively. The coefficient of correlation between plasma potassium and venous blood pH was (r_s = −0.44; p < 0.001). Scatterplots illustrating the association between plasma potassium concentrations, venous blood pH and serum inorganic phosphorus concentrations are shown in Figures 1C,D. Plasma potassium and sodium concentrations, as well as the potassium to sodium ratio was associated with alterations of posture/ability to stand (p < 0.001). Recumbent calves had significantly (p < 0.001) lower median cNa⁺ (108.7 mmol/L), higher cK⁺ (6.6 mmol/L), and a higher K⁺/Na⁺-ratio (5.9) than calves with unaltered posture/ability to stand (cNa⁺: 120.8 mmol/L; cK⁺: 4.6 mmol/L; K⁺/Na⁺-ratio: 3.8).

Acidemia (venous blood pH < 7.33) was present in 68 calves (57.6%), whereas alkalemia (venous blood pH > 7.37) was present in 27 calves (22.9%). Use of the Henderson-Hasselbalch approach for assessment of acid–base balance (by evaluating pCO₂, cHCO_3, and AG) indicated the presence of an acid–base derangement in 96 calves (81.4%), of which 68 calves were suffering from a mixed acid–base disorder (i.e., presence of more than one abnormality). Metabolic acidosis (cHCO₃⁻ < 23 mmol/L) was present in 71 calves (60.2%), whereas metabolic alkalosis (cHCO₃⁻ > 29 mmol/L) was present in 14 calves (11.9%). A venous pCO₂ acidosis (pCO₂ > 54 mm Hg) was present in 10 calves (8.5%), whereas a venous pCO₂ alkalosis (pCO₂ < 43.5 mm Hg) was present in 54 calves (45.8%). Increased values for AG (> 15 mEq/L) were detected in 54 calves (45.8%). Of the 71 calves with metabolic acidosis, 49 calves had increased anion gap (69%), 48 calves had decreased venous pCO₂ (67.6%) and one calf had increased venous pCO₂ (1.4%).

Use of the simplified strong ion approach to evaluating acid–base balance (by evaluating pCO₂, A_tot, SID₃, and SIG) indicated the presence of an acid–base derangement in 115 calves (97.5%), of which 89 calves had a mixed acid–base disorder. A strong ion (SID₃) acidosis (SID₃ < 38.3 mEq/L) was present in 86 calves (72.9%), whereas a strong ion (SID₃) alkalosis (SID₃ > 47.7 mEq/L) was present in 7 calves (5.9%). A buffer ion acidosis (A_tot > 21.2 mmol/L) was present in 34 calves (28.8%), whereas a buffer ion alkalosis (A_tot < 15.9 mmol/L) was detected in 23 calves (19.5%). A total of 59 calves (50.0%) had increased strong ion gap (< −3 mEq/L). Increased concentrations of unmeasured strong ions (USI ≤ 2 mmol/L) were present in 60 calves (50.8%). Among the 59 calves with increased SIG, 54 (91.5%) had also increased USI concentrations.

Spearman’s coefficients of correlation between clinicopathologic and acid–base variables are reported in Supplementary Table S1. Scatterplots illustrating the association between venous blood pH and SID₃, SID₅, SIG, SID_eff, A_tot, and pCO2 as independent variables of the simplified strong ion acid–base model are shown in Figure 2. The coefficients of correlation between venous blood pH and SID₃, SID₅, SIG, SID_eff, A_tot, and pCO2 were 0.33, 0.48, 0.67, 0.89, −0.43, and 0.45, respectively. Results of a stepwise forward linear regression analysis which aimed to predict venous blood pH and actual bicarbonate concentration by means of independent variables of the strong ion difference approach are reported in Table 2. Partial pressure of carbon dioxide was the most important predictor for actual bicarbonate concentration, whereas venous blood pH was best predicted by SIG and USI, respectively. As shown in Figure 3, SIG was associated with plasma L-lactate (r_s = −0.59) and serum inorganic phosphorus concentrations (r_s = −0.73).

In order to further assess associations with venous blood pH, plasma or serum concentrations of L-lactate, phosphorus, total protein, sodium, potassium as well as pCO₂ were entered into an additional multivariable stepwise linear regression model. Plasma chloride concentration as well as serum urea and creatinine concentrations were not entered because of their association with plasma sodium and serum inorganic phosphorus concentrations, respectively (Supplementary Table S1). Inorganic phosphorus, pCO₂, and serum total protein explained 30.2, 10.2, and 2.9% of the variation of venous blood pH, respectively (Table 3).

Preadmission treatments were reported to have been administered to 100 calves (84.7%), including analgesic drugs (61 calves), antibiotics (62 calves), and coccidiostatic agents (57 calves). Administration of intravenous fluids was documented in 19 calves (16.1%) at variable time points before hospital admission; however, none of these calves received intravenous fluids on the day of hospitalization.

During hospitalization, a total of 86 calves (72.9%) were treated with intravenous fluids containing 0.9% NaCl and if indicated hypertonic glucose (in concentrations of 10 to 40%) and sodium bicarbonate solutions (in concentrations of 1.4 to 8.4%). Oral electrolyte solutions were offered to 83 calves (70.3%). Intraruminal administration of sodium chloride or sodium bicarbonate (in amounts of 20 to 50 g) was performed in 31 (26.3%) and 22 calves (18.6%), respectively. Sodium salts were dissolved in variable volumes of tap water and administered as hypertonic solutions. All calves had access to water and a salt lick stone.

Considering pretreatment activities before hospitalization, coccidiostatic agents (diclazuril or toltrazuril) were administered orally in 42 (35.6%) of calves. Administration of parenteral antibiotics were continued or initiated in 79 calves (66.9%) due to presence of concurrent bacterial infection, suspected septicemia or fever. Additional supportive treatment consisted of administration of non-steroidal drugs in 112 calves (94.9%). Five calves received a blood transfusion, and 16 calves with persistent tenesmus were administered epidural anesthesia using 98% ethanol based on a previously published protocol (39). Surgical intervention due to umbilical herniation or signs of an acute abdominal emergency was performed in 5 calves.

The overall survival rate of calves of this study population was 60.2% (n = 71). The median duration (Q₁/Q₃) of hospitalization in these 71 calves was 15 (10/20) days. A total of 47 calves did not survive to hospital discharge of which 40 calves were euthanized and 7 calves died spontaneously. Reasons for euthanasia were not stated in most of the records. However, the presumed reason was most frequently ongoing depression and a concomitant lack of improvement of treatment measures (n = 22) and/or a massive deterioration of general condition (n = 17). A total of 12 calves were euthanized due to occurrence of an acute abdominal emergency. Presence of neurologic symptoms were only documented in one out 47 calves with a negative outcome, which had a postmortem diagnosis of cortical cerebral necrosis.

Associations between specific clinical finding and the outcome of treatment are reported in Table 4. Univariable binary logistic regression analysis indicated that impairments of ability to stand, presence of enophthalmos, increased skin tent duration, and poor body condition were associated with a higher risk for NO. Most clinicopathologic variables were similar for NO and PO calves. However, calves with NO had lower plasma sodium, chloride and ionized calcium concentrations, lower venous blood oxygen tension, as well as higher leukocyte counts and serum urea concentrations (Table 5). Measured and pH-corrected ionized calcium concentrations (calculated using Equation 9) had similar diagnostic accuracy. Therefore, measured ionized calcium concentration was used for multivariable modelling.

Multivariable logistic regression analysis indicated that mortality was associated with impairment of ability to stand, higher leukocyte concentration, and lower concentrations of plasma ionized calcium and serum albumin, as well as decreased plasma SID₅ (Table 6 and Figure 4)_. The area under the ROC curve of this model was 0.82 (95% CI: 0.74–0.90, p < 0.001) and the resulting sensitivity, specificity, PPV and NPV at the optimal probability cut-off of 0.35 was 78.7% (95% CI: 65.1–88.0%), 71.8% (95% CI: 60.5–81.0%), 64.9% (95% CI: 51.9–76.0%), and 83.6% (95% CI: 72.4–90.8%), respectively. Presence of concurrent health problems was not considered a confounder for this multivariable model.

Results of the classification tree analysis are illustrated by Figure 5. This analysis indicated that NO was associated with plasma ionized calcium concentrations <1.05 mmol/L on admission. In those calves, mortality was further associated with a poor or cachectic body condition. In contrast in calves with initial plasma calcium concentration > 1.05 mmol/L, initial leukocyte counts >16 × 10⁹ cells/L were associated with a higher risk for a NO. Based on predicted values, the area under the ROC curve of this model was 0.75 (95% CI: 0.65–0.85; p < 0.001). The sensitivity, specificity, PPV, NPV for predicting a negative outcome of this model was 59.6% (95% CI: 45.3–72.4%), 90.1% (95% CI: 81.0–95.1%), 80% (95% CI: 64.1–90%), and 77.1% (95% CI: 67–84.8%), respectively.