A prospective, blinded assessment of ammonia testing demonstrates low utility among front-line clinicians

Introduction

Although ammonia is essential to the pathophysiology of hepatic encephalopathy (HE), its levels cannot diagnose HE, do not correlate with the grade of HE, and are plagued by technical challenges.¹ Nevertheless, ammonia levels are routinely obtained in the evaluation of hospitalized patients.² We have advocated for quality improvement (QI) to limit testing.³ However, data is lacking regarding the reasons for, context and perceived value of ammonia testing among ordering clinicians. Additional data are needed to optimize a QI intervention aiming to curb overuse.

Methods:

We created an email alert generated each time an ammonia level was ordered (before the test resulted). We then contacted consecutive ordering providers while they were blinded to the pending ammonia level from 10/2020-12/2020. We surveyed the clinicians to determine their pre-test probability of/intention-to-treat HE and how the ammonia level would influence their decisions. We then followed the patients to determine their post-test diagnosis and changes in management. Bivariate analysis was performed with Fisher’s exact testing and multivariable logistic regression was conducted to compare the association of HE management (lactulose orders) and the clinician’s pre-test probability of HE adjusting for the ammonia level relative to the clinician’s pre-specified abnormal level. Our study was approved for quality improvement (HUM00188194).

Results:

All tests (100%) were ordered to evaluate for HE. Overall, 49/50 orders were placed in the emergency department (ED), by 25 physicians, 14 advance practice providers, and 11 triage nurses. Thirteen ammonia orders (26%) were made without having seen the patient. Thirteen (26%) patients did not have cirrhosis, 22 (44%) were ultimately diagnosed with HE (grades 0-3), and 31(62%) received lactulose. An ammonia level exceeding the clinician’s threshold for abnormal was associated with lactulose use (22/24 with high ammonia received lactulose vs 9/26 with low levels, p<0.001). Adjusting for the presence of cirrhosis and the clinician’s pretest probability of HE, an ammonia level greater than the clinician’s threshold for abnormal was neither associated with a lactulose order nor a HE diagnosis. Table 1 presents the patient and management details according to the clinician’s pre-test probability of HE. Overall, 36% of clinicians assigned a <25% pre-test probability of HE in a group with 1 diagnosis of HE (grade 1). In contrast, a >75% pre-test probability for HE was reported for 11 (22%) patients in the sample, 10 of them having a diagnosis of cirrhosis and 9 of whom were diagnosed with HE.

Table 1.

Clinical and Management Details According to Clinician’s Pre-Test Probability of Hepatic Encephalopathy (HE)

Clinician’s pre- test probability of HE	n (%)	Ammonia level “positive” for clinicians (±SD)	Ammonia test ordered before seeing patient	Cirrhosis (%)	Ammonia test result (±SD)	Provider who ordered test	Prior HE diagnosis	Lactulose administered	HE diagnosed
<25%	18 (36%)	67.8 (19.5)	6 (33.3%)	50%	33.1 (18.2)	Physician=11 APP=4 Nurse=3	4 (44%)	5 (55.6%)	1 (11%) Grade 1
25-75%	21 (41%)	69.5 (19.1)	3 (14.2%)	86%	82.1 (40.2)	Physician=11 APP=6 Nurse=4	16 (88.9%)	16 (88.9%)	8 (44%) 2 - Grade 3 5 – Grade 2 1 – Grade 1
>75%	11 (22%)	60.9 (21.2)	4 (36.3%)	91%	82.5 (48.6)	Physician=3 APP=4 Nurse=4	9 (90%)	9 (90%)	9 (90%) 1 - Grade 3 5 – Grade 2 3 – Grade 1 1 – Grade 0

HE: Hepatic encephalopathy, APP: Advance practice provider

Discussion:

Our study prospectively assessed the utility and impact of ammonia levels. We found that ammonia levels are ordered in low-probability patients leading to overuse of lactulose. When ordered in high-pretest probabilities, it does not influence decision making. Further, ammonia levels are often ordered before evaluating the patient and in non-cirrhotic patients

Ammonia at a crossroads

Data are mounting in support of ammonia testing stewardship. First, high ammonia cannot diagnose HE while low ammonia cannot exclude.¹ A recent study showed that neither ammonia orders nor their levels influenced the dose of lactulose provided or patient outcomes.⁴ Our data extend the field by providing a prospective quantification of the ammonia’s clinical value to clinicians blinded to the result. Second, ammonia levels are highly labile with unknown thresholds for abnormal values.⁵ In our study, we show both high variance in the clinician’s perception of the abnormal value as well as how their behavior was unaffected by the level regardless of their perceptions of its normality/abnormality. Third, a recent trial demonstrated that ammonia-level directed therapy does not change patient outcomes.⁶ In sum, these data show that ammonia should not be routinely tested and ought to be reserved for situations where it can change management. Potential interventions would a) eliminate ammonia testing in the emergency department or at least in triage, b) reserve testing for clinicians managing acute liver failure, valproic acid toxicity, or urea cycle disorders, or, the least stringent option, c) present clinicians with the data on test performance. In support of the latter, we have previously demonstrated substantial reductions in inappropriate testing by simply creating an alert displaying the expected test utility.⁷

Contextual factors

These data must be interpreted in the context of the study design. First, although our use of real-time surveys of result-blinded clinicians is innovative, it was conducted at a single center. Second, a study sample of 50 levels may be too small to detect rare but useful scenarios – i.e. evaluation of acute liver failure and valproic acid toxicity.

Conclusion

A quality improvement initiative aimed at reducing overuse is warranted, primarily in the emergency department and ideally targeted to reduce orders among the > 50% of clinicians who have low or high suspicion of HE.