Temperature Profile and Adverse Outcomes After Discharge From the Intensive Care Unit

We assembled a retrospective cohort study that included sequential adult patients who were discharged to a step-down unit after an acute admission to an ICU between 2015 and 2016. All included patients required mechanical ventilation for longer than 48 hours. We excluded from the cohort patients who were febrile at the time of ICU discharge or who were discharged with clear limitations on escalating care in the event of clinical deterioration; this group excluded patients who were discharged to palliative care. Patients recovering from acute brain injury or neurological injury were also excluded because of a significant risk of structural injury to their circadian pacemaker. Patients with a burn injury were excluded because they often have abnormal thermoregulation. Patients who met the inclusion criteria were sourced from the Royal Brisbane and Women’s Hospital MetaVision database (iMDsoft). This data set is well established and has been validated; it is used to collect real-time local data for inclusion in a national performance survey of critically ill patients through the Australia and New Zealand Intensive Care Centre for Outcome and Resource Evaluation. For patients with multiple ICU admissions within the study period, we included data from only the first ICU admission. The study received institutional ethical approval by the Research Ethics Committee of the Royal Brisbane and Women’s Hospital (LNR/2018/QRBW/48281).

The principal study outcomes were the number of MCAEs assessed within 7 days of ICU discharge (MCAE-7) and the number of RRSAs assessed within 7 days of ICU discharge (RRSA-7), based on the concept of 7 days as a practical follow-up period after ICU discharge. We collected RRSA data from the Rapid Response System database of the Safety and Quality Unit of the Royal Brisbane and Women’s Hospital. In addition to the time when the RRSA occurred, this database also records call criteria (cardiac arrest, threatened airway, breathing, circulation, neurology, worry) and the call outcome (death, palliative care, transfer to coronary care unit, transfer to ICU, or remained in general care area).

We assessed clinical adverse events as a secondary outcome and defined them a priori; relevant data were collected retrospectively from electronic clinical charts in the integrated electronic medical record (Cerner) (see Supplemental Table 1).¹⁸  An MCAE was defined as a combined end point of new-onset nosocomial pneumonia, confusion, or delirium; a deep venous thrombosis or pulmonary embolus; a myocardial infarction or new-onset arrhythmia; and a catheter-related bloodstream infection or a urinary tract infection. Conditions identified during chart review, such as delirium, were defined per clinicians’ records; although such definitions potentially increase the generalizability of the findings, they may limit the findings’ validity.

The principal independent variable was the final complete CBT-24, which represents temperature during the 24 hours before ICU discharge and was based on clinical temperatures recorded with an indwelling urinary catheter thermistor. Esophageal, tympanic, and axillary temperatures (in order of preference) were included only when bladder temperatures were not available. We assessed temperature profiles using the R statistical program and R package CircaCompare¹⁹ ; we compared patients’ CBT-24 measurements with those from simulated sex-matched controls, which we obtained from publicly available data.²⁰  We analyzed CBT profiles for the presence of circadian rhythmicity (P < .05); if rhythmicity was present we assessed the amplitude, the mesor or the midpoint of the oscillation, and the phase (P < .05). Patients with a body temperature higher than 38.5 °C were considered to be febrile.

Additional clinical data to describe the cohort included the Charlson Comorbidity Index,²¹  principal diagnosis at ICU admission (surgical, elective, medical, trauma, obstetric), hospital mortality, hospital and ICU lengths of stay, inotrope use, dialysis therapy, ileus, sepsis requiring antibiotic therapy at admission, and Acute Physiology and Chronic Health Evaluation (APACHE) II²²  and Sequential Organ Failure Assessment (SOFA)²³  scores both at ICU admission and upon discharge. We used APACHE II and SOFA scores at ICU discharge in an attempt to quantify the severity of illness at that time point, consistent with previous studies.^24,25  Medications administered within the 24 hours before ICU discharge were characterized by drug class; sedation on the day of discharge was summarized as a composite of major tranquilizers, narcotics, sedatives including α₂-agonists, and benzodiazepines.

We entered data into a purpose-built Microsoft Access research database (Microsoft Corp) and then analyzed the data using Stata 15 statistical software (StataCorp) to identify univariate associations and perform logistic regression modeling of the outcome of interest. After data linkage, the data used for the analysis were deidentified.

Sample size calculations were limited because no published data were avilable for the rate of circadian disruption or the frequency of RRSAs or MCAEs among patients at ICU discharge. We conservatively estimated the rate of RRSA or MCAE after ICU discharge to be 14%.¹²  Based on results of the z-test for a single proportion, 58 patients would be required if the absence of circadian temperature rhythmicity increased the RRSA rate to 25%. We aimed to enroll 300 patients in this initial pilot study, and considering the rule of thumb of 10 individuals with either an MCAE or an RRSA for each explanatory variable, the regression modeling should be able to support approximately 5 explanatory variables in the final model.²⁶  For descriptive statistics, we calculated absolute and relative frequencies for categorical parameters and characterized continuous parameters using median and interquartile range (IQR), though we provide an actual range when we consider it to be clinically relevant. We used the t, χ², Fisher exact, and Wilcoxon rank sum tests for inferential statistics, as appropriate. Results were considered to be statistically significant when P was .05 or less. We constructed forward, stepwise logistic models for RRSA-7 and MCAE-7. We considered variables for inclusion in the model when associations had a P of .25 or less in the univariate analysis or were clinically relevant. Goodness of fit was assessed with the Hosmer-Lemeshow statistic, whereas calibration was assessed on the basis of the area under the receiver operating characteristic curve. Sensitivity/specificity plots were derived for the probability of RRSA-7 and MCAE-7 at the median values of the covariants in the respective regression models.

We enrolled 291 patients in the study. Table 1 summarizes their demographic information, including age, ICU and hospital lengths of stay, case mix, comorbid conditions, and APACHE II and SOFA scores at admission and discharge. All patients received some form of active central nervous system analgesia or sedation within the 24 hours before ICU discharge. Of the 48 patients prescribed a sedative agent, only 8 (17%) received dexmedetomidine, whereas 30 (63%) received clonidine. Details about medication administration during the 24 hours before ICU discharge are summarized in Supplemental Table 2, and details about the Charlson comorbidities are summarized in Supplemental Table 3.

Data to analyze temperature profile were available from 289 patients. A median of 22 temperature measurements (IQR 11-24) were available for each of these patients. A minimum of 6 measures were required for cosinor analysis. In most patients, temperature had been measured at a range of sites; a breakdown of temperature measurement sources is detailed in Supplemental Table 4. Among these 289 patients, at ICU discharge, 191 (66%) had a temperature profile that was considered rhythmic by cosinor analysis (P < .05). The median mesor or central temperature of the oscillation was 37.4 °C (IQR 37.1-37.7), with a median amplitude of 0.28 °C (IQR 0.20-0.40) and median peak time of 4:25 PM (IQR 1:20 PM -7:13 PM ). Temperature peaked between 9 PM and 6 AM in 43 patients (23%). Representative temperature profiles are detailed in the Figure.

After ICU discharge, 100 RRSAs occurred; 64 patients (22%) had at least 1 RRSA. Median APACHE II and SOFA scores, along with additional RRSA details, are shown in Table 2. A breathing disturbance was the most common reason for an RRSA (n = 73 [73.0%]), although some patients were delirious or required cardiopulmonary resuscitation. With regard to RRSA outcomes, in most cases patients either remained in the general care area or were readmitted to the ICU; 3 (3.0%) patients died. Only 10 patients received palliative care, 3 at the time of the RRSA and 7 after it. Most RRSAs occurred during the first 7 days following ICU discharge.

A total of 192 patients (66%) experienced at least 1 clinical event during their stay in the general care area, with a median of 2 events (IQR, 1-16 events) per patient. The frequencies of each type of clinical event experienced during the stay in the step-down unit are summarized in Table 3. The median time after ICU discharge until the onset of a clinical event was 6 days (IQR, 0-103 days). Among all clinical events, 255 (43%) were MCAEs, the most common of which were delirium, pneumonia, thromboembolic disease, and urinary tract infection. More than 60% of patients experienced at least 1 MCAE. All clinical events that occurred after ICU discharge are summarized in Supplemental Table 5.

An RSSA-7 was associated with age, a diagnosis of sepsis at ICU admission, APACHE II and SOFA scores at ICU discharge, the APACHE III score at ICU admission, and the Charlson Comorbidity Index (Table 4). There was no association of RSSA-7 with CBT-24 rhythmicity at ICU discharge. Only age, sepsis at ICU admission, and Charlson Comorbidity Index were independent predictors of an RRSA-7. The predicted probability for sepsis at ICU admission for an RRSA-7 at the means of the covariates was 0.27 (95% CI, 0.18-0.35). Supplemental Figure 1 summarizes both sensitivity and specificity compared with probability cutoffs.

For MCAE-7, independent predictors were the absence of a rhythmic CBT-24, SOFA score at ICU discharge, Charlson comorbidity score, age, and male sex (Table 5). The predicted probability for the absence of a rhythmic temperature profile for an MCAE-7 at the means of the covariates was 0.60 (95% CI, 0.50-0.70). Supplemental Figure 2 summarizes sensitivity and specificity compared with probability cutoffs. Of the 147 MCAE-7 events, only 41 (28%) resulted in an RRSA-7.

Results of the TAOPID study showed that the absence of CBT-24 rhythmicity was an independent predictor of MCAE-7, as were illness severity at discharge, comorbidities, age, and sex. Such relationships were less robust beyond 7 days after ICU discharge, and the lack of CBT rhythmicity had limited effect in predicting an RRSA-7.

Rhythmicity of physiological processes is a key element of homeostasis and is optimally coordinated with geophysical time.²⁷  Many rhythmic processes are commonly disrupted in patients in an ICU^{1,3,4,28–34} ; the degree of disruption is associated with increasing illness severity.³¹  Recommendations are available for care routines and environments to facilitate circadian rhythms.^35,36  The effectiveness of therapies that attempt to normalize these timings and their benefit to critically ill patients have not been studied sufficiently. Our study demonstrates that, at least for temperature measurement, not all patients had a normal rhythmicity by the time of ICU discharge, which has implications for predicting MCAEs after ICU discharge.

Circadian rhythm dysfunction suggests an underlying abnormality or illness process.³⁷  Assessment of CRD can, however, be complex and expensive because of the need for repeated measurements of surrogate metabolic markers.³⁷  Body temperature is readily measurable in the ICU, but apart from defining fever, patterns of temperature profiles have not been well defined in critically ill populations.³⁸ 

Rapid response systems are generally reactive to alerts or parameters from various approaches to monitor vital signs,^39–42  though an RRSA does not always occur when a patient clinically deteriorates.⁴³  Many MCAEs may result in severe patient morbidity despite an RRSA not being called for all MCAEs. The performance of predictors for patient deterioration following ICU discharge are varied. Objective measures such as the level of C-reactive protein are better predictors than ICU admission type (medical, surgical, trauma), the timing of ICU discharge, SOFA score at ICU discharge, white cell count, or fibrinogen concentration.⁴⁴  Outreach systems for use after ICU discharge principally aim to preempt deterioration before abnormal physiological signs occur. Phenotypic profiling has utility in reducing ICU readmission rates within 48 hours but has not effected an overall change in patient outcome.⁴⁵  Existing scoring systems such as the National Early Warning Score have been predictive of clinical deterioration after an ICU stay,⁴⁶  and in particular for respiratory failure.^12,47  Our study did show consistent associations between SOFA score at discharge, MCAE-7, and RRSA-7. In the TAOPID study, we uniquely assessed CBT to objectively and simply predict patients’ progress after ICU discharge. We found utility for it, however, only within the early post-ICU period, similar to findings from previous predictive studies of clinical deterioration.¹²  This finding is relevant because it also defines a potential optimal time period for outreach services to patients who have been recently discharged from the ICU.

The relation between MCAE and the patient’s sex may reflect a complex relationship between clinical outcomes, age, and sex⁴⁸ ; women aged younger than 50 years have a lower adjusted mortality rate than men, who receive more intense treatment and have higher ICU readmission rates.^49–51  Hormonal differences may also influence responses to sepsis and hypoxia.^52,53  Illness specificity may be important, as adjusted mortality rates differ between the sexes for illnesses such as chronic obstructive pulmonary disease and ischemic heart disease.⁴⁸  Men may have higher morbidity than women after trauma.⁴⁸  Such effects may be reflected in the influence of male sex on the regression model, increasing the likelihood of an adverse clinical event.

Pragmatically, we used all temperature assessment modalities that were available within the last 24 hours of the ICU stay, as this process represents clinical reality. Measurements of CBT obtained with bladder catheter thermistors are not always used in the last days of an ICU stay, when the intensity and invasiveness of clinical monitoring may be reduced. Most patients, however, did have enough clinical measurements available to allow us to determine CBT rhythmicity. The temperature profile may be truncated with the use of all clinically available routes of temperature measurement. As such, in this study we focused assessment on the dichotomized presence of temperature rhythmicity as this is more likely than amplitude or phase to be preserved across measurement methods. Using regularly available data, we were able to predict patient deterioration. The limited comparison data available are from 24-hour temperature profiles of healthy patients with public domain data stratified by sex only. Despite a lack of age matching, normal temperature profiles are still rhythmic across all age groups. We expected that temperature profiles from similarly aged healthy people, used as controls for comparison, would not be arrhythmic or phase reversed, as is the case with temperature profiles in many ICU patients. The TAOPID study was an initial exploration of CBT-24 as a simple, readily available, and objective measure of the cyclic character of temperature profile at ICU discharge and its potential clinical utility. When CBT is collected as part of an electronic clinical information system, its inclusion in scoring systems that are used for predicting clinical deterioration after ICU discharge could provide an additional assessment when stratifying the need for post-ICU follow-up.

The TAOPID study supports the use of readily available CBT-24 measurements along with illness severity at ICU discharge to assist in stratifying the risk of subsequent deterioration of patients during care in a step-down unit.