The diagnostic accuracy of clinical examination for estimating cardiac index in critically ill patients: the Simple Intensive Care Studies-I

We conducted the prospective, observational, single-centre SICS-I following a pre-published protocol and statistical analysis plan (SAP; clinicaltrials.gov; NCT02912624) [17]. All consecutive patients admitted to the intensive care unit (ICU) of the University Medical Center Groningen (UMCG) between 27 March 2015 and 22 July 2017 were considered eligible. We included patients who were aged 18 years or older, had an unplanned ICU admission and were expected to stay for at least 24 h. We excluded patients if their ICU admission was planned preoperatively, if acquiring research data interfered with clinical care (e.g., mechanical circulatory support) and if informed consent was not provided (e.g., refusal, serious language barrier). In unresponsive patients, informed consent was first obtained from the legal representatives. Consent for use of the study data was asked at a later time if the patient recovered consciousness. If the patient died before consent was obtained, the study data was used, and the legal representatives were informed on the study. The study was approved by the local institutional review board (METc M15.168207).

All patients were included within the first 24 h of their ICU admission and underwent clinical examination, immediately followed by CCUS. Patients were included by medical research interns and PhD students who had received a focused CCUS training course given by experienced cardiologist-intensivists (protocol in Supplements 1) [16]. These researchers were not involved in patient care, and their findings were not revealed to the patients’ caregivers.

All clinical examinations were standardised and cut-off values for abnormal clinical signs predefined in the protocol (clinicaltrials.gov; NCT02912624). We recorded in total 19 clinical signs per patient (e-Table 1). Respiratory rate, heart rate and rhythm, arterial blood pressures and central venous pressures were recorded from the bedside monitor. Patients were auscultated for the presence of cardiac murmurs and crepitations. Clinical signs reflecting organ perfusion were obtained from the three organs readily accessible to clinical examination: cerebral (mental status), renal (urine output) and skin perfusion (capillary refill time (CRT), central-to-peripheral temperature difference (ΔTc–p) and skin mottling). Mental status was assessed with the AVPU scale, which consists of the categories ‘Alert’, ‘responsive to Voice’, ‘responsive to Pain’ and ‘Unresponsive’ and was not scored in patients who were receiving sedative drugs or who were admitted after a trauma. Urine output was scored 1 and 6 h prior to the clinical examination, adjusted for body weight, and considered decreased if < 0.5 mL kg⁻¹ h⁻¹. CRT was the time for skin colour to fully return after applying firm pressure at the sternum, index finger, and knee for 15 s and considered prolonged if > 4.5 s [18]. ΔTc–p was the difference between central temperature measured by a bladder thermistor catheter and peripheral temperature measured by a skin probe on the big toe and dorsum of the foot and considered abnormal if > 7 °C [19]. The degree of skin mottling was rated at the knee according to a score from 0 to 5, where 0–1 was regarded as mild, 2–3 as moderate and 4–5 as severe mottling [20].

Cardiac index was measured by transthoracic echocardiography using the Vivid-S6 system (General Electric, Horton, Norway) with cardiac probe M3S or M4S, and with default cardiac imaging setting. The parasternal long axis (PLAX) was used to measure the left ventricular outflow tract (LVOT) diameter. In the apical five chamber (AP5CH) view, a pulse wave Doppler signal in the LVOT was used to measure the velocity time integral (VTI): three VTIs were traced when the heart rhythm was regular and eight VTIs when the heart rhythm was irregular. Cardiac output was calculated using an established formula and was adjusted for body surface area, i.e. cardiac index, because it allowed us to compare patients with different body dimensions [21]. We tested the diagnostic accuracy of clinical signs on four frequently used cut-off values: below 2.2 and 2.5. and above 4.0 and 4.5 L min⁻¹ m⁻² [10, 11, 22, 23].

All CCUS images were validated and each cardiac index measured by experts from an independent imaging core laboratory (Groningen Imaging Core Laboratory, www.​g-icl.​com). These experts were blinded for the clinical examination findings. The most-often used modality for measuring cardiac index in the literature currently is the transpulmonary thermodilution method [24, 25]. We chose CCUS because it is non-invasive, its use is advocated by guidelines, and ensured high-quality measurements throughout validation by experts who followed recent guidelines [26].

No previous research has studied clinical examination in a cohort of consecutively included critically ill patients. Therefore, we estimated our sample size based on the number of acute ICU admissions annually (n = 1500) and estimated that half would fulfil the inclusion criteria. When our sample size exceeded 1000 patients, we calculated a potentially detectable difference using CRT as an example; we were able to detect a 0.10 L min⁻¹ m⁻² increase in cardiac index for each second of CRT increase with a power of 100% and alpha of 0.015 [17]. Missing clinical examination values were imputed using multiple imputations (20 times) as these were considered missing at random. Regression coefficients of our final models were averaged using Rubin’s formula [27]. Central venous pressure was missing in 822 (76%) patients and was therefore excluded from the analyses and imputations. Following our SAP, we also imputed missing cardiac index values based on validated LVOT diameters and VTIs (Supplements 2; sensitivity analyses).

The aims of our analyses were twofold: first, we conducted a least-squares linear regression analysis to identify the clinical examination findings that were independently associated with cardiac index as a continuous variable. Second, we calculated diagnostic test accuracies for each clinical sign and conducted multivariable logistic regression analyses to determine which combined clinical signs were independently associated with a low (< 2.2 and < 2.5 L min⁻¹ m⁻²) or high (> 4.0 and > 4.5 L min⁻¹ m⁻²) cardiac index.

A total of 1212 patients were eligible, of which 137 patients were not included because the desired CCUS window was obstructed by chest drains, wounds, emphysema, or prone positioning (n = 80); routine clinical care such as acute surgery or angiography did not allow patient inclusion (n = 40); or other reasons such as colonisation with multiresistant bacteria (n = 17), leaving 1075 patients for inclusion. Of these, 292 (27%) patients were excluded because one or both of the CCUS views were of insufficient quality as scored by the independent echocardiography core laboratory, resulting in a total of 783 (73%) patients included in the current analyses (Fig. 1). One-third of the patients were admitted after acute or complicated surgery, and the most common admission diagnoses were of cardiovascular or respiratory origin (Table 1). The median time from ICU admission to inclusion was 15 h (IQR 8–20 h).

The mean cardiac index was 2.65 ± 0.93 L min⁻¹ m⁻². Table 2 shows that 12 clinical examination findings differed significantly over five cardiac index categories. Clinical signs reflecting normal skin, renal and cerebral perfusion were present in 36 (5%) patients. Abnormal clinical signs were observed in 598 (76%) patients for skin perfusion, 536 (68%) patients for renal perfusion, and 204 (26%) patients for cerebral perfusion. Both cerebral and renal perfusion were abnormal in 141 (18%) patients; skin, renal and cerebral perfusion were abnormal in 103 (13%) patients.

Univariable analyses in the overall population showed that heart rate, systolic blood pressure and all clinical signs reflecting abnormal organ perfusion were statistically significantly associated with cardiac index (supplements; e-Table 2). The strongest association was found for heart rate with an R² of 0.15. At the mean heart rate of 87 beats per minute, the mean cardiac index was 2.1 L min⁻¹ m⁻² with an individual 95% CI that ranged from 0.87 to 4.1 L min⁻¹∙m⁻² (e-Fig. 1), reflecting the inaccuracy of a univariable prediction. In multivariable linear regression seven clinical signs, i.e., respiratory rate, heart rate and rhythm, systolic and diastolic blood pressure, ΔTc–p, and CRT were independently associated with cardiac index (e-Table 3). The presence of cardiac murmurs had a p value > 0.015 and was statistically significant in less than 80 of the 100 bootstrap replications. The multivariable model had an R² of 0.30 (98.5% CI 0.26–0.37). Other model diagnostics and sensitivity analyses are presented in the supplements (e-Figs. 2–4 and e-Tables 4, 5).

Clinical signs are important warning signals of a deteriorating patient condition, but cannot reliably indicate whether the cardiac index is low, normal or high, or the underlying cause. Our results show both the relevance and limitations of clinical examination: five clinical signs had a specificity above 90%, yet the PPV is much more pertinent than specificity in the clinical process, as it reflects the probability that a low CI is present given that the clinical signs indicate hypoperfusion. The PPV of single clinical signs (Fig. 2) is too low in the majority of patients. PPVs above 90% were observed in a minority of patients with several abnormal clinical signs (Fig. 3). It is unacceptable for a diagnostic test to miss a low cardiac index in one-quarter of cases, or even a larger proportion when diagnosing a high cardiac index (e-Table 6). If physicians immediately perform CCUS, they could diagnose cardiac failure within a few minutes and perform the adequate supportive and therapeutic measures. In the acute setting, this is superior to the collection of urine over 1 h.

The results of our study imply that physicians should not rely solely on clinical signs for their decision making. Although the clinical signs atrial fibrillation, systolic blood pressure below 90 mmHg, altered consciousness, CRT of  > 4.5 s at the sternum and skin mottling over the knee make a low cardiac index very likely, we advocate ultrasonography for additional evaluation in these patients. For patients with normal clinical signs additional evaluation should be performed on indication. Therewith, we verify two important statements of the current circulatory shock guidelines for which no evidence base existed (i.e., ‘best practice’) [1].

The clinical examination findings measured in the SICS-I were prone to confounding by interventions. The administration of noradrenaline may have influenced both cardiac index and organ perfusion variables, which is why we added it as a confounder in our multivariable models. The observational design of our study  does not allow for conclusions if the measured cardiac index is sufficient for the individual patient. CCUS is focused on obtaining few potentially important measurements and therewith valve pathologies can be overseen. Additional CCUS measures such as ejection fraction and end-diastolic volumes are needed to distinguish between a compensating or a failing heart, but this was beyond the scope of our current research question.

The SICS-I included acutely ill patients admitted to the ICU of a tertiary referral hospital. Compared to others who selected patients with sepsis or myocardial infarction, we studied the value of clinical examination in a large, unselected cohort. Since PPV and NPV values are dependent on the baseline prevalence of low cardiac index, we selected all consecutive patients to maximise representativeness of our cohort to the ICU. However, our study requires external validation before its findings can be applied to other ICUs. Furthermore, our study did not test the inter- and intra-observer agreement of cardiac index measured with CCUS; we will address the inter-observer agreement (researcher versus expert) separately.