Peripheral Perfusion Index Trajectory to Predict 28-Day Mortality in Patients With Sepsis in the Intensive Care Unit

Background and rationale

Sepsis and septic shock remain leading causes of intensive care unit (ICU) mortality. Resuscitation is commonly guided by macrohemodynamic targets such as mean arterial pressure (MAP), urine output, and lactate clearance. However, restoration of macrocirculatory targets does not guarantee adequate microcirculatory perfusion. This macro-micro discordance, sometimes referred to as occult or hidden hypoperfusion, has been associated with progressive organ dysfunction and worse outcomes, yet it is not captured by routine monitoring.

The peripheral perfusion index (PI) is derived from the standard pulse oximetry waveform and represents the ratio of pulsatile to non-pulsatile signal. It is continuous, non-invasive, requires no additional equipment beyond routine monitoring, and reflects peripheral perfusion. Most available evidence has evaluated PI at a single time point, which may not capture the dynamic nature of resuscitation. Whether the trajectory of PI over the first 24 hours provides prognostic information beyond established severity scores has not been adequately established.

Objectives

Primary objective: to determine whether the trajectory of the peripheral perfusion index during the first 24 hours of ICU admission independently predicts 28-day all-cause mortality in adults with sepsis or septic shock.

Secondary objectives: (1) to quantify the incremental predictive value of PI trajectory when added to a reference model comprising SOFA score and arterial lactate; (2) to compare the discriminatory performance of PI, capillary refill time (CRT), skin mottling score, and lactate within the same cohort; (3) to define a macro-micro discordance phenotype (MAP at target despite persistently impaired peripheral perfusion) and evaluate its association with 28-day mortality; (4) to develop a TRIPOD-compliant prediction model and evaluate it in an independent center.

Study design and setting

Prospective, observational, two-center cohort study conducted in adult ICUs. This is a non-interventional study. All diagnostic and therapeutic decisions, including fluid administration, vasopressor titration, antimicrobial therapy, and organ support, are made exclusively by the treating clinical team according to institutional practice and current guidelines. No study-specific intervention, additional blood sampling, or invasive procedure is performed. PI values obtained for research purposes are not disclosed to the treating team, in order to prevent measurement-driven changes in management.

Population

Consecutive adults (≥18 years) meeting Sepsis-3 criteria for sepsis or septic shock, enrolled within the first 6 hours of ICU admission. Key exclusion criteria include conditions precluding valid peripheral perfusion measurement (severe peripheral arterial disease, limb ischemia or amputation, extensive edema or burns, local trauma, severe Raynaud phenomenon), ICU stay of 24 hours or longer at another facility prior to admission, anticipated withdrawal of life-sustaining treatment within 24 hours, pregnancy, persistent tremor or agitation precluding measurement, and concurrent enrollment in an interventional study. Each patient is enrolled only once, at the first eligible ICU admission (index admission). If a patient is readmitted to the ICU during follow-up, no new record is created and the patient remains a single observation unit; index marker measurements are not repeated at readmission.

Measurements and schedule

Peripheral perfusion is assessed at ICU admission (hour 0) and at 6 and 24 hours using a standardized bedside protocol. PI is measured with the patient supine, the limb at heart level, nail polish removed, after a 2-minute stabilization period; three consecutive readings are obtained from the right index finger (or a predefined alternative site) and averaged. CRT is measured using a standardized technique with firm pressure applied to the ventral surface of the index finger for 10 seconds using a glass slide, with the return of normal skin color timed; values above 3 seconds are considered abnormal. Skin mottling score (0-5) is assessed over the knee area. Arterial lactate is recorded at hours 0 and 24 from routine blood gas sampling, and lactate clearance is calculated. Contextual variables recorded concurrently with each measurement include MAP, norepinephrine-equivalent vasopressor dose, skin temperature, and sedation depth. Severity of illness is characterized using SOFA and APACHE II scores, and comorbidity burden using the Charlson Comorbidity Index. Time to first appropriate antimicrobial therapy, appropriateness of empirical therapy, and source control are recorded as treatment-related confounders. Inter-observer agreement for PI, CRT, and mottling score is assessed in a predefined subsample using duplicate measurements.

Outcomes

The primary outcome is 28-day all-cause mortality. Secondary outcomes include ICU and in-hospital mortality, 90-day mortality, vasopressor-free days at day 28, duration of mechanical ventilation, new organ failure, shock reversal, and ICU and hospital length of stay. All outcomes are defined over a 28-day calendar window starting at enrollment rather than over the index ICU stay, and encompass all ICU episodes occurring within that window.

Sample size

The target sample size is approximately 400 prospectively enrolled patients, with a stopping rule of 400 patients or 100 deaths, whichever occurs first. This was derived using criteria for developing clinical prediction models, targeting approximately 15 events per candidate predictor with an assumed 28-day mortality of 25%, and is compatible with the sample required to detect a clinically meaningful difference in c-index between the reference model and the model incorporating PI trajectory.

Statistical analysis

The association between PI trajectory and 28-day mortality will be assessed using multivariable Cox proportional hazards regression, with PI modeled both continuously (using restricted cubic splines to accommodate non-linearity) and dichotomized at a predefined threshold. Longitudinal PI behavior will be modeled using linear mixed-effects models. Incremental predictive value over the reference model (SOFA and lactate) will be evaluated using discrimination (Harrell c-index, time-dependent AUC, DeLong test), reclassification (continuous NRI, IDI), calibration (calibration slope and intercept, observed-to-expected ratio), and decision curve analysis. Missing data will be handled using multiple imputation by chained equations, with complete-case analysis as a sensitivity analysis. Model development will incorporate penalized regression and shrinkage, with internal validation by bootstrapping. Prespecified sensitivity analyses will adjust for vasopressor dose, temperature, antimicrobial timing and appropriateness, and source control. Patients who die before the 24-hour assessment are retained in the cohort and are not excluded; their deaths are counted as primary outcome events. Measurements not obtained because of death are treated as truncated by death rather than missing, and are not multiply imputed. Analyses involving trajectory-based variables are therefore conducted as a prespecified 24-hour landmark analysis restricted to patients alive at 24 hours, with follow-up starting at that landmark, whereas the primary analysis based on baseline measurements includes the entire cohort.

Development and validation centers

The prediction model will be developed in the cohort enrolled at the development center and subsequently evaluated, without any modification of model coefficients, in the cohort enrolled at the independent validation center. The assignment of centers to development and validation roles was defined prior to data collection and is not modified thereafter. External evaluation will report discrimination, calibration slope and intercept, observed-to-expected ratio, and clinical net benefit. Reporting will follow the TRIPOD statement, with STARD and STROBE as applicable.