Metabolic acidosis (MA) is a major physiological derangement associated with mortality. ¹ There is no specific treatment for MA, beyond treating its cause. However, the administration of buffering solutions to correct MA may be common practice in the intensive care unit (ICU). The rationale for such therapy is that, independent of its cause, MA with acidaemia impairs the performance of the cardiovascular system ^{2, 3} and should be corrected. To reduce this perceived cardiovascular dysfunction, sodium bicarbonate is the most commonly used agent. ^{4, 5} However, until recently there have been little or no data on its effectiveness. ^{6, 7}

In 2018, Jaber and colleagues ⁸ reported the findings of a multicentre, open-label, randomised controlled trial of the efficacy of intravenous sodium bicarbonate for critically ill patients with severe MA (BICAR-ICU trial). The trial investigators found that the administration of sodium bicarbonate reduced the use of renal replacement therapy (RRT) in the ICU, and decreased 28-day mortality in the subgroup of patients with acute kidney injury Stage 2 and 3. These findings have created renewed interest in the use of sodium bicarbonate for MA.

However, it is unclear how many patients such an intervention would apply to and what the outcome of such patients in the setting of usual care might be. Moreover, for the purpose of designing and powering a trial, it is unknown whether investigators should apply more moderate criteria for MA to achieve more efficient patient selection. In particular, patients with moderate MA may also have a high mortality risk and a much greater prevalence than severe MA. This larger population would then enable the conduct of a trial with greater external validity and with recruitment that is more efficient.

Accordingly, the aim of the study was to measure the incidence, prevalence, characteristics and outcomes of patients with both severe MA (using the BICAR-ICU trial criteria) and moderate MA (using new criteria) in Australian and New Zealand ICUs.
 

We conducted a retrospective cohort study using the Australian and New Zealand Intensive Care Society (ANZICS) Adult Patient Database (APD). The ANZICS-APD is a high quality binational database that encompasses more than 90% (n = 193) of all ICUs in both countries.

We screened admission records of these 193 ICUs from January 2008 through to December 2018. We excluded the following admissions: ICU admission for palliative end-of-life care or purpose of organ donation, readmission to the ICU in the same hospital stay, pH and arterial partial pressure of carbon dioxide (PaCO₂) not measured simultaneously, base excess suggestive of data errors (< -30 mmol/L or > 30 mmol/L), and missing data for outcome at hospital discharge.

The Alfred Health Ethics Committee, Melbourne, Australia, approved the study protocol (No. 775/19).
 

Variables included baseline characteristics, laboratory data and urine output on ICU admission day and treatments during the ICU stay of patients. In addition, the database contains ICU and hospital admission data and data on survival outcomes at ICU and hospital discharge.
 

The collection methodology for blood gas analysis (BGA) data in the ANZICS-APD was changed in 2016 (Online Appendix). The system change was considered likely to cause a reduction in the recorded incidence and prevalence of acidosis compared with the previous period or the true rates. Thus, we treated the period from 2008 to 2016 and the period from 2017 to 2018 as separate and non-comparable periods. Moreover, the criteria for BICAR-ICU-type MA were studied in 2017 and 2018 because of lack of complete relevant data on lactate in the 2008–2016 period. Therefore, all comparisons between the two criteria were performed in this latter period. However, we were able to report the incidence and prevalence of moderate MA separately for the two periods.
 

The following criteria were used to identify patients with early MA.
 

All laboratory data collected in the ANZICS-APD were obtained during the first 24 hours in the ICU and the BGA data were selected to deliver the values producing the highest score for the Acute Physiology and Chronic Health Evaluation (APACHE) III-j prognostic system (Online Appendix)

BICAR-ICU criteria set No. 1 (early severe MA criteria):

Novel ANZICS Criteria set No. 2 (moderate early MA criteria):

Incidence, prevalence, patient characteristics and the outcomes of patients with MA by the two criteria were summarised using descriptive statistics. Group comparisons were performed using 𝑥² or Fisher exact test for categorical variables, Student t test for normally distributed data and Wilcoxon rank sum test otherwise, with results presented as number (percentage), mean (standard deviation) and median (interquartile range) respectively. Diagnostic accuracy of the two criteria for hospital mortality was determined by calculating sensitivity, specificity, positive predictive value, and negative predictive value.

To evaluate whether MA diagnosed by the two criteria was an independent predictor of mortality, hierarchical logistic regression models were fitted adjusting for key factors. These factors included APACHE III-j score without the acidosis component, sex, chronic conditions, planned ICU admission, limitations of treatment orders at the ICU admission, ICU diagnosis categories of APACHE III-j, medical or surgical, potassium and PaCO₂, with patients nested within sites and sites treated as a random effect. 

We further explored incidence and prevalence of patients with MA, as well as hospital mortality of patients with and without MA stratified by year. To investigate the change in hospital mortality over time by moderate MA, logistic regression models were fitted with main effects for year of admission, MA along with their two-way interaction and adjusting for APACHE III-j risk of death, with ICU site treated as a random effect.

We further explored patient characteristics and outcomes in patients with MA diagnosed by both criteria after excluding chronic renal failure, ketoacidosis, and possible exogenous acidosis (patients with overdose) so as to simulate the patient screening in the BICAR-ICU trial in Australian and New Zealand ICUs.

We did not impute any missing values and all available data were analysed. We performed all analyses using R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria.) or SAS software version 9.4 (SAS Institute, Cary, NC, USA). Given the magnitude of the dataset, in order to more closely align statistical and clinical significance, a two-sided P  value of 0.01 was chosen to indicate statistical significance.
 

Among the 1 513 836 ICU admissions registered in the ANZICS-APD, 437 749 records met the pre-set exclusion criteria, leaving 1 076 087 patients for analysis (Figure 1). Of these, 244 740 were admitted to the ICU from 2017 to 2018. In this period, patients with MA by the BICAR-ICU criteria accounted only for 1.5% (1350/87 110; 157 630 patients were excluded due to missing data) of the patients in the ICU, whereas 8.4% (20 679/244 740) of patients had at least moderate MA (P < 0.001).

Table 1 shows the characteristics, the interventions provided in ICU, and the outcomes of all patients and of patients with MA according to the two sets of criteria. Hypercapnia of PaCO₂ > 45 mmHg was observed in 25.3% (61 967/244 740) of all patients. In addition, the rates of the hypercapnia were 54.6% (24 859/45 538) and 56.2% (6754/12 024) of patients, with pH < 7.30 and < 7.20 respectively. According to protocol, such patients were excluded.

Patients with MA by either criterion had higher mortality than those without MA.Hospital mortality was higher in patients with MA diagnosed by the BICAR-ICU criteria than moderate MA criteria (P < 0.001). Sensitivity, specificity, positive predictive value, and negative predictive value for hospital mortality of the BICAR-ICU criteria and moderate MA criteria were 8.4%, 99.1%, 48.3%, 91.7% and 21.7%, 92.8%, 21.5%, 92.8% respectively. Logistic regression analyses revealed that early MA by both criteria was independently associated with hospital mortality (early severe MA: adjusted odds ratio [aOR], 2.431; 95% CI, 2.098–2.817; P < 0.001; early moderate MA: aOR, 1.542; 95% CI, 1.464–1.623; P < 0.001).
 

Table 2 shows the characteristics and outcomes of patients who met the eligibility criteria of the BICAR-ICU trial, after excluding end stage renal failure, ketoacidosis, and possible exogenous toxin acidosis. These patients were compared with patients with moderate MA after excluding these conditions.
 

Changes in the annual incidence and prevalence of moderate MA are shown in Figure 2. For moderate MA, the average incidence rates during the period from 2008 to 2016 and the period from 2017 to 2018 were 411.5 per million per year and 349.2 per million per year respectively. While for early severe MA, the incidence was 39.5 per million per year in 2018. Annual unadjusted hospital mortality of patients with and without moderate MA as well as adjusted annual odds ratio for the hospital mortality are shown in Figure 3. The P value for interaction effect between moderate MA and year of admission was 0.016 for the period 2008 to 2016; however, the P value was 0.78 in the latter period from 2017 to 2018.
 

Using the ANZICS-APD, we found that the diagnostic criteria for MA used in the BICAR-ICU trial applied to only for 1.5% of critically ill patients, and that hospital mortality for such patients was 48.3%. In contrast, patients with moderate MA accounted for 8.4% of the Australian and New Zealand ICU population and still carried a significant hospital mortality of 21.5%. In the latest 2 years of the study period, using a different methodology for BGA inclusion, the mortality of moderate MA appeared to be increasing.

Day 28 mortality in the BICAR-ICU study was 49.1%. ⁸ In comparison, hospital mortality in the Australian and New Zealand ICU patients who met the eligibility criteria in the BICAR-ICU trial was 51.8%. This is despite some minor differences in the criteria applied. These findings support the external validity of the mortality rate reported in the BICAR-ICU trial.

RRT was used in 34.9% (68/195) and 51.5% (100/194) of patients in the intervention and control groups of BICAR-ICU trial respectively, with an overall value of 43.2%. In comparison, RRT use in the Australian and New Zealand ICU patients was 41.6%, supporting the external validity of renal outcomes of the BICAR-ICU trial. However, the prevalence of severe MA using the BICAR-ICU criteria was quite low in Australian and New Zealand ICUs (1.5%).

The BICAR-ICU trial was designed following observations from five French ICUs by Jung and colleagues. ¹ In 2011, they reported that the prevalence of metabolic or mixed severe acidaemia of pH < 7.20 within the first 24 hours of ICU admission was 6%. However, only patients with severe MA without hypercapnia were enrolled in the BICAR-ICU study. The study by Jung et al ¹ did not report on the ratio of metabolic and mixed acidosis. However, our study found that 56.2% of Australian and New Zealand patients who had pH < 7.20 also had PaCO₂ > 45 mmHg, suggesting that the prevalence of isolated metabolic acidosis in the previous study may have been about 2.6% if adjusted for hypercapnia. This value would be > 70% higher than found in our study.

A recent guideline on the diagnosis and treatment of MA recommended that the pH not be used alone to identify such patients. ⁹ Smith and colleagues ¹⁰ suggested that < -4 mmol/L of the initial base excess in ICU is a useful predictor for mortality. In the current study, 8.4% of patients in the Australian and New Zealand ICUs were diagnosed with MA by the moderate MA criteria on their ICU admission day based on pH, a base excess lower than -4 mmol/L, and the lack of hypercapnia. Hospital mortality among such patients was as high as that of septic shock in Australian and New Zealand ICUs, ¹¹ and appeared to be increasing in 2018.
 

In both the BICAR-ICU trial and the current study, patients with severe MA had very high mortality. However, in Australian and New Zealand ICU patients, in the first 24 hours, the prevalence and incidence of the BICAR-ICU trial criteria were low. At about 40 patients per million per year and assuming that, as in previous similar studies, only 30 of 190 ICUs would participate in such a trial and that anticipated recruitment would be 30% of eligible patients as in other trials ¹² of acutely ill patients in Australia and New Zealand, this would translate into only about 50 patients randomised per year. As about 550 patients would be required to have a > 80% power to detect a 25% relative risk reduction in mortality from a baseline of 48%, such a trial would require 11 years and would not be feasible in Australia and New Zealand. The same would apply if RRT were the primary outcome.

In contrast, for moderate MA, a 25% decrease in mortality with sodium bicarbonate therapy from 21.5% at α = 0.05 would require the randomisation of 1750 patients to achieve > 80% power. Assuming the observed incidence of 400 per million per year and the same anticipated unit participation and recruitment rate, 540 patients would likely be recruited per year making such a trial feasible in 3 years. If choosing RRT as the outcome, 1250 patients would have to be randomised, making such a trial even more feasible.
 

The use of a binational database including more a million patients made it possible to provide precise demographics of patients with MA in real-world ICUs. The large sample size enabled us to describe the small population defined by the BICAR-ICU criteria, and to explore subpopulations with or without hypercapnia. The results are widely generalisable to Australian and New Zealand ICUs. They allow reasonably accurate estimates of trial requirements for the design and powering of future randomised controlled trials of an intervention such as sodium bicarbonate.

This study has several limitations. First, its retrospective observational design cannot differentiate whether the observed outcomes were induced by the MA or by other conditions. Second, about one-third of records were excluded mainly due to missing data in diagnostic variables, which might have caused a degree of selection bias. For example, patients who did not have an arterial line might have also had missing data on BGA. Third, the database used in the present study included only laboratory data on the day of ICU admission. Therefore, we could not assess the trends of such variables. Fourth, information on specific treatment such as administration of intravenous sodium bicarbonate was not available.
 

BICAR-ICU criteria could identify patients with MA carrying high mortality; however, the prevalence of such severe MA was quite low in Australian and New Zealand ICUs, making an interventional trial in such patients non-feasible. Nevertheless, defining MA with broader criteria could identify patients who were still at high risk of death and had a much higher incidence, making a trial in such patients feasible over a period of 3 years in Australia and New Zealand alone. These observations have important implications for the design and powering of a future randomised controlled trial of sodium bicarbonate in these patients.