Cerebrospinal fluid quinolinic acid is strongly associated with delirium and mortality in hip-fracture patients

Abstract

BACKGROUNDThe kynurenine pathway (KP) has been identified as a potential mediator linking acute illness to cognitive dysfunction by generating neuroactive metabolites in response to inflammation. Delirium (acute confusion) is a common complication of acute illness and is associated with increased risk of dementia and mortality. However, the molecular mechanisms underlying delirium, particularly in relation to the KP, remain elusive.METHODSWe undertook a multicenter observational study with 586 hospitalized patients (248 with delirium) and investigated associations between delirium and KP metabolites measured in cerebrospinal fluid (CSF) and serum by targeted metabolomics. We also explored associations between KP metabolites and markers of neuronal damage and 1-year mortality.RESULTSIn delirium, we found concentrations of the neurotoxic metabolite quinolinic acid in CSF (CSF-QA) (OR 2.26 [1.78, 2.87], P < 0.001) to be increased and also found increases in several other KP metabolites in serum and CSF. In addition, CSF-QA was associated with the neuronal damage marker neurofilament light chain (NfL) (β 0.43, P < 0.001) and was a strong predictor of 1-year mortality (HR 4.35 [2.93, 6.45] for CSF-QA ≥ 100 nmol/L, P < 0.001). The associations between CSF-QA and delirium, neuronal damage, and mortality remained highly significant following adjustment for confounders and multiple comparisons.CONCLUSIONOur data identified how systemic inflammation, neurotoxicity, and delirium are strongly linked via the KP and should inform future delirium prevention and treatment clinical trials that target enzymes of the KP.FUNDINGNorwegian Health Association and South-Eastern Norway Regional Health Authorities.

Keywords: Dementia; Inflammation; Metabolism; Neurological disorders; Psychiatric diseases.

Figure 1. Overview of study design and included patients.

We undertook a multicenter observational study with 586 hospitalized patients and investigated associations between delirium and KP metabolites measured in CSF and serum by targeted metabolomics. We also explored associations between KP metabolites and markers of neuronal damage and 1-year mortality. Patients from 2009 to 2019 were included, and CSF samples were available from 450 hip-fracture patients (224 with and 226 without delirium), 24 participants from an independent cohort of patients with medical delirium, and 112 cognitively unimpaired adults scheduled for elective surgery under spinal anesthesia. Serum samples collected at the same time as CSF were available from 338 patients. Delirium was assessed daily according to DSM-5 criteria and based on a standardized procedure. See Supplemental Appendix.

Figure 2. Trp, KP metabolites, and delirium in the hip-fracture cohort.

Unadjusted and adjusted logistic regression with delirium as the outcome in the hip-fracture cohort (224 with delirium and 182 without delirium; 44 with SSD excluded from analysis). Serum metabolites (A, n = 213) and CSF metabolites (B, n = 406) were both associated with delirium, although CSF metabolites (B) were in general more strongly associated with delirium and remained so in adjusted analyses. The strongest association was with CSF QA. Covariates were age, sex, renal function (GFR), ASA score (I–II versus III–IV), and IQCODE (≥ 3.44 versus < 3.44). Serum Trp was untransformed; CSF Trp, Kyn, HK, Pic, and QA were transformed using the inverse square root transformation. All other metabolites were log transformed. Following transformation, the metabolites were standardized to a mean of 0 and an SD of 1. Inverse square root transformations generally provide somewhat lower ORs than log transformations. *Q < 0.05; **Q < 0.001 (Q values are P values adjusted for multiple comparisons so that Q < 0.05 indicates significance after adjustment for multiple comparisons).

Figure 3. QA concentrations in CSF and delirium.

(A) The form of the effect size of the QA-delirium association using logistic regression is depicted on the nmol/L scale (n = 406, 224 with delirium and 182 without delirium, 44 with SSD excluded). The marginal predictions from logistic regression using transformed CSF-QA (B) have been back-transformed to the original scale of CSF-QA in nmol/L in unadjusted (red) and adjusted (green) analyses, with age, sex, GFR, ASA score (I–II versus III–IV), and IQCODE (≥3.44 versus <3.44) as covariates. In A, CSF-QA values greater than 500 nmol/L (n = 4) have been left out for illustrative purposes, but were included in the statistical analyses. In the background, one can see the highly skewed distribution that has been transformed to approximate normality (B). On the transformed scale (B), a 1 SD increase in CSF-QA gives an OR of 2.26 (unadjusted) and 1.79 (adjusted) for delirium (Figure 1). However, this is highly nonlinear on the nmol/L scale (A), where the per unit effect of CSF-QA on the probability (Pr) of delirium decreases incrementally as CSF-QA concentrations increase. This was confirmed using per 50 nmol/L and per 100 nmol/L quantitative cutoffs as predictors of delirium (data not shown).

Figure 4. QA, NMDAR agonists, and potential neurotoxicity.

Potentially neurotoxic concentrations of CSF-QA, although observed in a minority of delirium patients, were much more frequent in delirium (A). Glutamate and aspartate, like QA, stimulate the NMDAR. Although there was no significant difference in CSF glutamate and aspartate concentrations in hip-fracture patients by delirium presence, the overall tendency was for the highest concentrations to occur in cognitively unimpaired adults and the lowest concentrations in medical delirium (B). QA was significantly associated with the neuronal injury marker NfL in hip-fracture patients using standardized linear regression (C and D), also adjusted for age, sex, renal function, and cognitive impairment (IQCODE ≥ 3.44), ASA score (I–II versus III–IV), and delirium status (standardized linear regression with an interaction). Data shown in A and B included 406 hip-fracture patients (excluding SSD), 112 cognitively unimpaired adults, and 24 patients with medical delirium. Data shown in C–F included all hip-fracture patients (SSD patients classified as not delirium in adjusted analyses). However, 16 cases did not have NfL measured, and thus, there were 434 patients with hip fracture in this analysis (not excluding SSD cases). The association between QA and NfL was stronger in patients with cognitive impairment (E) and high ASA scores (F).

Figure 5. CSF-QA and 1-year survival in patients with hip fracture.

Using a cutoff of 100 nmol/L, there was a clear survival benefit to patients with lower QA concentrations in univariate analyses (A) where the hazard for death was the highest in the initial period following the hip fracture (B) S(t), survival function. In unadjusted analyses, a 1 SD change in CSF-QA on the transformed scale resulted in a reduction in survival as illustrated in C (hazard ratio of 2.37); this was somewhat attenuated in adjusted analyses as illustrated in D (hazard ratio of 1.76). Of note, back-transforming the mean of QA to nmol/L results in 48.8 nmol/L, where –1 SD is 25.7 and +1 SD is 126.0. Kaplan-Meier curve in A and HR with 95% CI from Cox regression. B–D were estimated using a standardized parametric survival analysis where the baseline hazard was estimated using restricted cubic splines. n = 450; 99 events.