Key words intensive care unit - mortality prediction - adrenal gland - spleen
Introduction
The CT hypoperfusion complex [1 ]
[2 ]
[3 ]
[4 ]
[5 ] includes a number of imaging features that can be seen when pronounced hypoperfusion
is present in the case of organ conditions that are usually life-threatening. At the
end of the 1980 s, the CT hypoperfusion complex was initially described in polytraumatized
children. Perfusion changes in the gastrointestinal tract, pancreas, and spleen were
seen in cases of hypovolemic shock [6 ]. The clinical signs of shock are often not reliable. Therefore, particularly in
critical patients without shock symptoms, an image-based morphological indicator promised
information about a decrease in a patient’s overall hemodynamic condition and radiological
identification of “patients at risk”. Consequently, individual studies on ICU patients
examining the classic “flat cava” sign of the vena cava [7 ] as well as additional organs were published. In 2020, Elst et al. stated that, even
though it has been a long time since the sign was initially observed, there is still
no consensus on the frequency or clinical relevance of this sign [8 ]. Reduced enhancement of the spleen [9 ], increased enhancement of the adrenal glands [10 ], and increased enhancement of the aorta [11 ]
[12 ] are often seen in the CT hypoperfusion complex. Qualitative descriptions are based
on ROI shy;(region of interest)-based analyses of the enhancement pattern of individual
organs or measurements of organ and vascular diameters resulting in an elevated mortality
risk in the group comparison. Arbitrary thresholds [13 ]
[14 ] were used for classification purposes and allowed statements regarding individual
risk to a limited degree [14 ].
The study presented here focuses on intensive care patients. In the clinical routine
in radiology, we noticed hyperdense adrenal glands on contrast-enhanced CT in many
of these patients with some of these patients dying in the following days. As a result
of the different approaches used in previous studies regarding the selection of contrast
phases and organs, we decided to perform a systematic comparison of enhancement patterns
in the arterial and portal venous phases on the basis of ROI-based measurements [15 ]
[16 ]. We focused on the adrenal glands, the spleen, and the large abdominal vessels.
In ROC (receiver operating characteristics) analyses, the portal venous contrast phase
showed the greatest relationship between organ density values and mortality. In this
phase organ densities already have a substantial predictive value regarding the risk
of death within the first three days after imaging [15 ]. However, the quality of the prediction increases significantly when the organ density
values are viewed in relation to one another or to the large abdominal vessels [16 ]. In a pilot study including 133 patients, the most significant parameter for short-term
mortality prediction in critically ill patients was the adrenal-to-spleen ratio. If
the threshold of 1.41 was exceeded, the probability of death was high [16 ] with the best prediction results being achieved regarding mortality within 72 hours
of imaging.
The organ parameters could be easily and quickly measured and could also be effectively
reproduced. Therefore, the adrenal-to-spleen ratio seemed to be a suitable image-based
screening instrument for the identification of short-term life-threatening conditions
in critically ill patients. However, this predictor has not yet been examined with
respect to stability in a large and different study cohort. That was the goal of the
present study. In the new cohort, any reason for imaging was taken into consideration
to thus recruit a representative group of intensive care patients needing radiological
imaging. The goal was to examine whether the threshold calculated in the pilot study
from the adrenal-to-spleen ratio also allows good discrimination between deceased
patients and survivors in the new cohort. We also examined whether the new threshold
calculated based on the new cohort differed significantly from the threshold calculated
in the pilot study and if there was a relevant difference in the prediction quality
of the two thresholds regarding short-term survival.
Materials and Methods
Patients and study design
The ethics committee approved the retrospective study (EK 414 092 019). We searched
the radiology information system (RIS) at our facility for intensive care patients
who underwent contrast-enhanced CT of the abdomen in the portal venous phase between
June and August 2020. Patients with pathologies of the spleen and adrenal glands,
a lack of documentation, and death due to termination of life-sustaining measures
were excluded. 290 intensive care patients underwent an abdominal CT scan. 371 CT
scans of 203 intensive care patients (127 men, age: 68.1 ± 14.4 years old) met the
inclusion criteria ([Fig. 1 ]).
Fig. 1 Study flowchart with numbers/reasons of included/excluded patients. † = number of
deceased patients.
CT image acquisition and post-processing
The patients were examined with a 128- or 192-slice scanner (Somatom Definition AS
or Edge and Force, Siemens, Forchheim, Germany). All examinations were performed with
110 kV and automatic tube current modulation (CAREDose 4 D, Siemens). The contrast
agent was administered by injecting 1 ml/kg body weight of a non-ionic iodized contrast
agent (Ultravist 370 mg/kg, Bayer Schering Pharma, Berlin, Germany or Solutrast 370 mg/kg,
Bracco Imaging Germany GmbH, Constance, Germany) with a flow rate of 3–4 ml/s, followed
by 50 ml of a saline solution. The portal venous phase was documented 70 seconds after
the start of contrast injection and stored in our image archiving and communication
system (IMPAX Agfa HealthCare, Bonn, Germany).
Image evaluation
A radiologist with 4.5 years of experience in abdominal CT evaluated the CT examinations.
The image data was evaluated on a PACS workstation (IMPAX Agfa HealthCare, Bonn, Germany).
The contrast enhancement of the adrenal glands and the spleen was analyzed on images
in the portal venous phase in 3-mm slices. For the quantitative analysis, the Hounsfield
units (HU) were determined and averaged preferably in the center of each adrenal gland
based on ROI measurements in axial slices. All measurements were performed using highly
magnified images to avoid adjacent fat. The HU of the spleen was determined by placing
three circular ROIs (2.0 cm2 ) on three different axial planes through the cranial, middle, and caudal third of
the spleen in the portal venous phase [9 ]. Hypodense triangular regions on the periphery of the spleen caused by small splenic
infarctions were not included in the measurement. The averaged HU value was used for
the data analysis. The dimensionless adrenal-to-spleen ratio was defined by the average
HU of the adrenal glands divided by the average HU of the spleen.
Statistical analysis
The data analysis was performed using MedCalc 19.6.1 (MedCalc Software bvba, Ostende,
Belgium). Characteristics of the study population were given as means and standard
deviations for continuous variables. After calculation of the adrenal-to-spleen ratio
from the organ density values, the patients were classified based on the threshold
defined in pilot studies for 72-hour (short-term) mortality. In addition, the corresponding
Matthews correlation coefficients (MCCs) were calculated based on the data points
and the assigned binary confusion matrices of the current ROC curve for the 72-hour
mortality, and the current threshold for the adrenal-to-spleen ratio, the area under
the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative
predictive value (NPV), as well as the positive and negative likelihood ratios were
calculated for the maximum MCC [17 ]. The MCC was used as the maximization criterion in both the pilot study and the
current study since a predictor reaches a high MCC in both balanced and unbalanced
datasets when it correctly classifies most cases under consideration of all categories
of the confusion matrix [18 ]
[19 ]. The area under the precision-recall curve (PRC) was additionally calculated. The
PRC as a performance metric should take into account the influence of the inclusion
of survivors with an adrenal-to-spleen ratio below the threshold in the new cohort
and the resulting potentially relevant increase in the imbalance between deceased
patients and survivors. As a result of the increase in specificity, this can result
in an ROC curve with an increased area under the ROC curve and thus a presumably better
test result. The positive predictive value (= precision) addressed solely in the PRC
is important for evaluating whether patients with an adrenal-to-spleen ratio above
the threshold actually died. Only an improvement in the PRC indicated an actual improvement
in test performance. The ROC, PRC, relative chance, and relative risk are shown for
the best mortality prediction [20 ]
[21 ]. Statistical significance was defined as p < 0.05.
Results
Patient outcome
20 intensive care patients (9.9 %) died within the first 24 hours, 38 patients (18.7 %)
within 48 hours, and 49 patients (24.1 %) within 72 hours of imaging. For all time
intervals, there was a significant difference in the group comparison between patients
who died and patients who survived for the density of the spleen and adrenal glands
as well as the adrenal-to-spleen ratio, with the group of survivors on average having
an adrenal-to-spleen ratio of less than or equal to 1 while those who died had a value
of approx. 2.4–2.6 ([Table 1 ], [Fig. 2 ]).
Table 1
Two-sample t-test results (equal sample variance)/Welch’s t-test (unequal sample variance)
for 24-, 48- and 72-hour mortality.
Survivors vs. Deceased patients
Adrenal glands (HU)
Spleen (HU)
Adrenal-to-spleen ratio
24-hour mortality
84.95 ± 24.96 vs.
113.60 ± 34.42
(Welch’s t -Test; p = 0.0016)
90.79 ± 29.42 vs. 54.61 ± 20.71
(t -Test; p < 0.0001)
1.00 ± 0.50 vs.
2.58 ± 1.47
(Welch’s t -Test; p = 0.0001)
48-hour mortality
83.72 ± 22.88 vs.
109.70 ± 38.86
(Welch’s t -Test; p = 0.0002)
93.07 ± 28.00 vs.
57.48 ± 26.70
(t -Test; p < 0.0001)
0.94 ± 0.36 vs.
2.34 ± 1.29
(Welch’s t -Test; p < 0.0001)
72-hour mortality
82.92 ± 22.49 vs.
108.52 ± 35.95
(Welch’s t -Test; p < 0.0001)
94.47 ± 27.23 vs. 57.03 ± 25.65
(t -Test; p < 0.0001)
0.89 ± 0.16 vs.
2.32 ± 1.24
(Welch’s t -Test; p < 0.0001)
Fig. 2 a Initial abdominal portal venous CT scan of a 53-year-old patient after gastrectomy.
The white arrow points to the adrenal gland (averaged adrenal density values: ≈ 72);
the white circle lies in the spleen (averaged spleen density values: ≈ 72; adrenal-to-spleen
ratio: ≈ 1). The patient survived the 72 hours following imaging. b Repeated abdominal portal venous CT scan of the same patient at a later time point
due to vital endangerment. The white arrow points to the adrenal gland (averaged adrenal
density values: ≈ 107); the white circle lies in the spleen (averaged spleen density
values: ≈ 24; adrenal-to-spleen ratio: ≈ 4.46). The patient died of septic multiorgan
failure within the next 72 hours.
Validation of the prediction quality of the ROC-based adrenal-to-spleen ratio threshold
determined in the pilot study regarding the 72-hour mortality prediction
Using the threshold for the adrenal-to-spleen ratio determined in the pilot study
(1.41) to classify the data resulted in 96 % of intensive care patients (= accuracy)
being classified correctly in the confusion matrix with regard to adrenal-to-spleen
ratio over or under the threshold as well survival or death within 72 hours of imaging
(sensitivity = 81.63 %, specificity = 99.07 %, PPV = 93.00 %, NPV = 97.30 %, +L = 87.62,
-L = 0.19). As in the pilot study, the adrenal-to-spleen ratio provided good discriminatory
power for mortality prediction (AUC = 0.97) with a high sensitivity and specificity.
At a threshold of 1.41, 13 of 371 datasets were classified incorrectly (false positive:
3, false negative: 10).
Accuracy of current ROC-based organ thresholds for the adrenal gland and spleen regarding
72-hour mortality prediction
With a low MCC and a very low sensitivity, the adrenal gland threshold on its own
is not suitable as a prognostic factor ([Table 2 ]). In general, the spleen has significantly better values for the individual predictive
parameters but has insufficient sensitivity.
Table 2
Results of the ROC analysis of the studied mortality predictors. Cut-off values/ratios
for 72-hour mortality prediction in the portal venous phase. AUC = area under the
curve. SENS = sensitivity. SPEC = specificity. PPV = positive predictive value. NPV = negative
predictive value. +LR = positive likelihood ratio. -LR = negative likelihood ratio.
MCC = Matthews correlation coefficient. PRAUC = area under the precision-recall curve.
ROC analysis for 72-hour mortality predictors
Predictor
Cut-off
AUC
SENS (%)
SPEC
(%)
PPV
(%)
NPV
(%)
+LR
-LR
p-value
MCC
Accuracy
PRAUC
Adrenal glands
> 142 HU
0.730
16.33
98.76
66.70
88.60
13.14
0.85
< 0.0001
0.29
87.88 %
0.262
Spleen
≤ 43 HU
0.870
44.9
99.69
95.70
92.20
144.57
0.55
< 0.0001
0.63
92.45 %
0.704
Adrenal-to-spleen ratio
> 1.37
0.974
83.67
99.07
93.20
97.60
89.81
0.16
< 0.0001
0.86
97.03 %
0.922
Accuracy of the current ROC-based adrenal-to-spleen ratio threshold regarding 72-hour
mortality prediction
The adrenal-to-spleen ratio threshold determined based on the current cohort (1.37)
deviates only minimally from the threshold determined in the pilot study and has a
slightly higher sensitivity of 83.67 % (95 % CI: 70.3–92.7) and an identical specificity
of 99.07 % (95 % CI: 97.3–99.8) for the risk of death within the first 3 days after
imaging (AUC = 0.97, p < 0.0001, MCC = 0.87) ([Fig. 3 ], [Table 2 ]).
Fig. 3 Comparison of ROC curves regarding short-term mortality prediction: the adrenal-to-spleen
ratio is best suited for short-term mortality prediction (AUC = 0.974). The area under
the curve as a measure of diagnostic accuracy is significantly larger for the adrenal-to-spleen
ratio than for the organ density values of the adrenals (ΔAUC = 0.244; p < 0.0001)
and spleen (ΔAUC = 0.104; p = 0.0013).
The positive likelihood quotient was 89.55 (95 % CI: 28.9–278.9), the negative likelihood
quotient was 0.16 (95 % CI: 0.09–0.3) with a resulting post-test probability of approximately
93 % with a pretest probability of 24.1 % (72-hour mortality). The relative chance
was 544.95 (95 % CI: 139.01–2136.45); the relative risk was 38.09 (95 % CI: 19.12–75.87).
At a threshold of 1.37, 11 of 371 datasets were classified incorrectly (false positive:
3, false negative: 8).
PRC analysis for 72-hour mortality
Using an adrenal-to-spleen ratio threshold of 1.37, the PRC showed ([Fig. 4 ]) a high sensitivity and a high PPV (PRAUC = 0.922) with a correct prediction rate
of 93.2 %.
Fig. 4 Precision-recall curves (PRC) for the adrenal-to-spleen ratio (PRAUC = 0.922). For
a threshold of > 1.37, the best ratio of precision (93.2 %) to sensitivity (83.67 %)
is obtained.
Discussion
The goal of the study was to evaluate the suitability of the adrenal-to-spleen ratio
as an image-based screening parameter for predicting short-term mortality. The prediction
quality of the adrenal-to-spleen ratio threshold determined in the pilot study and
the newly determined adrenal-to-spleen ratio threshold was determined and compared
based on a new cohort of intensive care patients. In contrast to the cohort in the
pilot study with vital indication and triphasic emergency CT, intensive care patients
without vital endangerment were also included in our study since an image-based prognostic
parameter should be applicable for every intensive care patient.
Our results show that the thresholds of different cohorts deviate only slightly from
one another, and the adrenal-to-spleen ratio can effectively predict short-term mortality
even when including patients without vital endangerment.
The discrimination between survivors and deceased patients using the thresholds is
highly accurate regardless of whether the threshold is from our own cohort or a different
cohort. Therefore, the adrenal-to-spleen ratio threshold determined in the study collective
(1.37) can identify critically ill patients with a high risk of death within 3 days
of CT examination with high sensitivity (83 %) and accuracy (93 %). The threshold
determined in the pilot study (1.41) yields almost identical values with a sensitivity
of 82 % and an accuracy of 93 %.
The spleen is a highly perfused organ that serves as a reservoir for thrombocytes
and red blood cells [15 ]. In life-threatening conditions, activation of the sympathetic nervous system and
humoral effects on the supply vessels and the spleen’s own connective tissue seem
to reduce arterial inflow and to increase venous outflow [6 ]. These drainage effects can be reflected in a decrease in density values after contrast
administration as shown by our results from the comparison of means and the ROC analysis.
Patients with sepsis or septic organ failure make up the largest percentage (23 %)
and one in three of these patients died. Di Serafina et al. emphasize that, even with
comparable CT findings, the pathogenic mechanisms of hypotensive shock and septic
shock are very different [22 ]. In general, elevated adrenal density values are associated with activation of the
hypothalamic-pituitary-adrenal axis. However, it should be assumed that the hypothesis
regarding greater enhancement of the adrenal glands due to increased secretion of
catecholamines [10 ] in stressful situations is not sufficient. Therefore, Peng et al. observed a specific
enhancement pattern of the adrenal glands on contrast-enhanced two-phase CT in a collective
of 194 critically ill patients with septic shock [23 ]. Significantly less washout was seen in the central zone of the adrenal gland than
in the peripheral zone in the arterial phase. Given the high mortality in both groups,
the incidence of this sign was significantly higher in the septic shock group (almost
30 %) than in the hemorrhagic shock group (0 %). The authors postulated that specific
pathophysiological changes occur during septic shock, with disrupted microcirculation
of the adrenal gland in the arterial phase having a greater effect on the central
zone of the adrenal gland with less enhancement than in the peripheral zone. Moreover,
swelling of the adrenal gland is a sign of functional and structural damage. It is
unclear whether the adrenal gland changes are an epiphenomenon of critical circulatory
situations or have a direct connection to these.
A mortality factor that must be considered is the administration of exogenous catecholamines
in intensive care patients since the vasoconstrictive effect can result in nonocclusive
mesenteric ischemia in a relevant portion of intensive care patients, which is associated
with a high mortality rate.
A limitation of the study is its retrospective design. A prospective study with additional
inclusion of laboratory parameters, intensive care medication, and organ function
scores would therefore be desirable. A further limitation is the measurement accuracy.
The relatively small adrenal glands were digitally magnified in an attempt to minimize
inaccuracies. Nonetheless, adjacent fat tissue and partial volume effects could have
affected the measurements. Using a combination of spleen and adrenal measurements
to determine the organ ratio can also result in error propagation. We tried to limit
this by averaging spleen measurements and excluding triangular, wedge-shaped, and
inhomogeneous regions. The measurements in this study were performed by a single radiologist
since the intraclass correlation coefficient of previous studies (0.90) [15 ]
[16 ] showed very high agreement among evaluators.
The adrenal-to-spleen ratio takes into consideration the opposite enhancement patterns
with decreasing density of the spleen and increasing density of the adrenal glands.
The organ enhancement ratio has excellent discriminatory power for differentiating
between deceased patients and survivors (AUC: 0.97) regarding the 3-day mortality
rate. It can be assumed that a presumable cause, i. e., the condition of the intensive
care patient, and effect, i. e., the death of the patient, are much more closely related
in this short interval after imaging than at later points in time. The MCC was selected
as a classifier because it is an adequate statistic in unbalanced datasets with a
significantly higher number of patients who survived than died. The inclusion of patients
requiring intensive care but without a vital indication and with low test results,
i. e., in our case a low adrenal-to-spleen ratio, can generally significantly improve
an ROC curve without increasing the sensitivity or the positive predictive value of
the evaluated parameter. However, the use of the PRC as a further performance metric
resulted in the adrenal-to-spleen ratio in the current cohort also being characterized
by a good positive predictive value.
In a substantial percentage of intensive care patients, opposite enhancement patterns
of the adrenal glands and the spleen can be observed in the portal venous CT phase
and are associated with an increased short-term mortality rate.
Quantitative analysis of the adrenal-to-spleen ratio in critically ill patients provides
retrospective information about the mortality and survival risk in the follow-up period
since there is an extremely high risk of death within the next 72 hours if a threshold
of approximately 1.4 is exceeded.
The adrenal-to-spleen ratio can be easily and quickly determined and can be reliably
reproduced.
This radiological parameter can be used with reliable reproducibility in prospective
studies to identify patients with an elevated risk of short-term death and can be
included in therapeutic considerations of intensive care physicians.
