PDF icon Download PDF
Open Access
CC BY 4.0 · J Neuroanaesth Crit Care
DOI: 10.1055/s-0045-1810607
Case Report

Exploratory Modeling of Intraoperative Co-oximetry Data for Predicting Hemodynamic Trends in a Thalassemic Patient: A Pilot Case

Kunal K. Sharma 
1   Department of Neuroanesthesia Super-speciality Cell, Indira Gandhi Medical College, Shimla, Himachal Pradesh, India
,
Kotipi Rajagopal M. Reddy
2   Department of Neuroanesthesia and Neurocritical Care, National Institute of Mental Health and Neurosciences (NIMHANS), Bengaluru, Karnataka, India
› Author Affiliations
› Further Information
Also available ateRefMedOne
 
 PDF Download Permissions and Reprints

Abstract

Background

Thalassemia minor presents unique challenges in intraoperative hemodynamic management due to chronically low hemoglobin and altered oxygen-carrying capacity. While pleth variability index (PVi) is an established surrogate of volume responsiveness, its behavior in hemoglobinopathies remains underexplored. This pilot case investigates the relationship between PVi and co-oximetry-derived parameters to assess perfusion trends when plethysmographic signals may be unreliable.

Methods

A 36-year-old female with thalassemia minor undergoing parasagittal craniotomy was monitored using Masimo Radical-7 co-oximeter (spectral hemoglobin [SpHb], perfusion index [Pi], spectrophotometric oxygen content [SpOC], PVi) and GE Centricity hemodynamic records. Data were time-synchronized using POSIX conversion and down-sampled to 1-minute intervals. Multivariate regression and LOESS (locally estimated scatterplot smoothing) curve fitting were used to explore relationships among SpHb, SpOC, Pi, and PVi.

Results

Regression modeling yielded:

PVi = 0.45 × SpHb + 0.15 × SpOC + 1.57 × Pi − 1.04,

with an adjusted R2 of 0.92. Pi emerged as the strongest predictor of PVi. LOESS plots revealed nonlinear associations, especially between PVi and Pi. SpHb showed minimal early-phase variability, consistent with limited responsiveness to acute blood loss.

Conclusion

This exploratory model highlights physiologically grounded inter-variable behavior of co-oximetry parameters in thalassemia. It provides a foundation for redundant trend monitoring and decision support during neurosurgery in patients with hematologic disorders.


 

Keywords

thalassemia - co-oximetry - hemodynamic - predictive modeling - monitoring - intraoperative

Introduction

Thalassemia is a hematological disease that is characterized by absence or reduction of β globin chain caused by mutations on 11p15.4, which forms the underlying etiology for this disease.[1] Intra-operative optimal transfusion of blood products is required to prevent tissue hypoxemia of vital organs and maintain adequate cerebral oxygenation.[2] Published literature depicts the rare association of this co-existing disease with cerebral sinus thrombosis, Moyamoya disease, etc.[3] [4] There is evidence that links posterior reversible encephalopathy to intraoperative hypoperfusion during neurosurgery in patients with hemoglobinopathies.[5] [6]

The use of real-time hemoglobin index monitoring for neurosurgery is yet to gain a strong foothold in the current neuro-anesthesia practice. Recent trends of predictive algorithms in neuroanesthesia are gaining a foothold in modern neuroscience.[7] [8] Therefore, in this pilot case, we investigate the interdependence of displayed parameters on Masimo Radical-7 co-oximeter, to test the sensitivity and early warning potential of these co-oximetry trends using an exploratory modeling.

The purpose of the model is not to replace pleth variability index (PVi), but rather to (1) explore the interdependence between co-oximetry-derived variables (spectral hemoglobin [SpHb], spectrophotometric oxygen content [SpOC], perfusion index [Pi]) and PVi in a patient with altered hemoglobin physiology (thalassemia minor); (2) quantify the contribution of each parameter toward perfusion trends using real-time, noninvasive monitors; (3) build redundancy into physiological monitoring by identifying surrogate variables that can reflect perfusion dynamics in situations where PVi fails (e.g., poor pleth waveform due to vasoconstriction, electrosurgical interference, or hardware malfunction).

A 36-year-old female with lateral ventricular lesion having thalassemia minor trait was scheduled for fronto-parietal parasagittal craniotomy and resection of a lateral ventricular lesion via transcallosal approach. The informed consent of patient was taken as per the CARE (CAse REport) guidelines from EQUATOR network (Enhancing the Quality and Transparency of health research Network).

Thalassemia minor usually results in mild anemia manifesting as slightly lower hemoglobin levels. As such, the patient's baseline SpHb was 10.3 g/dL, which was on the lower end. Hypoxia or hypotension could jeopardize brain function and recovery; therefore, critical assessment of tissue hypoxemia by adequately judging the hemoglobin value in real-time was required for this patient. We collected the hemodynamic and ventilatory data from the digital anesthesia chart (Centricity High Acuity Anesthesia software from GE Healthcare). However, the co-oximetry data were obtained by Masimo Radical-7 co-oximeter. We monitored SpHb, SpOC, PVi, and Pi through this device.

Intraoperative blood loss of 600 mL occurred during the resection of the lesion, for which 390 mL of packed red blood cells was transfused. The patient was extubated at the end of the surgery and had a positive balance of 820 mL.


 

Methods

Masimo Radical-7 co-oximeter was used to record SpHb, Pi, PVi, and SpOC in a patient with thalassemia minor undergoing neurosurgical intervention. In parallel, continuous hemodynamic monitoring was performed using the GE Centricity anesthesia record system, which captured data at 1-minute intervals.

Given the asynchronous logging intervals between the two monitoring systems, a time alignment protocol was implemented. First, the time discrepancy between the Masimo Radical-7 co-oximeter and the Centricity system was calculated based on a manual timestamp match, and this correction factor was applied to the Masimo co-oximetry data. All timestamps were standardized to the POSIX format for consistent handling. Following this, overlapping time windows were extracted, and the Masimo Radical-7 co-oximeter data were down-sampled to 1-minute intervals using a fixed interval selector to match the Centricity timestamps.

To explore the inter-relationships between variables, scatter plots were constructed and fitted with LOESS curves rather than linear regression, as LOESS better captures nonlinear associations within local data subsets. Data points with low PVi (<1) were retained to reflect full physiological variability, especially since in a single-case study such points offer insight into extreme responses.

Multivariate linear regression was performed in Python using the “statsmodels” library to estimate the contribution of SpHb, SpOC, and Pi to changes in PVi. The model was expressed as:

PVi = α . (SpHb) + β. (SpOC) + γ. (Pi) + δ

where α, β, and γ are coefficients that represent how much each variable contributes to changes in PVi. However, δ is a constant offset which accounts for baseline perfusion values.

Intraoperative data demonstrated distinct trends across co-oximetry and perfusion variables during key surgical events. At the onset of craniotomy-induced blood loss, PVi declined from 14 to 9, while SpHb showed a modest drop from 10.3 to 9.9 g/dL. SpOC remained relatively stable at 13 mL/dL until after blood transfusion, when it increased to 14 mL/dL. Pi increased sharply from 0.4 to 1.7 during this period, before normalizing following hemodynamic resuscitation.

These qualitative data led to the assumption that SpHb contributed in a smaller proportion to changes in PVi (since the change in SpHb was minimal). SpOC showed only a small response to the blood loss, so it likely has a smaller coefficient. While Pi is more responsive and therefore it will likely have a larger coefficient.

The model's performance was assessed using adjusted R2, F-statistic, and 95% confidence intervals for each coefficient. The data analysis was performed in conjunction with the hemodynamic data for modeling. We used the above-mentioned time-stamped data points in Python software, utilizing pandas for data handling and statsmodels for ordinary least squares regression. The time points where Pi < 1.0 were excluded from construction as they would have adversely affected the downstream modeling by compromising the accuracy of SpHb.


 

Results

The dataset included 23 time-synchronized data points after preprocessing. PVi ranged from 0.5 to 14, Pi from 0.4 to 1.7, SpHb from 9.5 to 10.5 g/dL, and SpOC from 13 to 14 mL/dL. Retention of low PVi values allowed for the examination of perfusion extremes, particularly during episodes of intraoperative blood loss and subsequent transfusion. The multivariate regression yielded the following model:

PVi = 0.45 × SpHb + 0.15 × SpOC + 1.57 × Pi − 1.04

This model demonstrated an adjusted R2 = 0.92, indicating the model to be a good fit for data, as there is a strong relationship between the predictors and the outcome. The model's overall statistical significance was confirmed with an F-statistic of 75.2 (p < 0.001). The 95% confidence intervals were respectively as follows: SpHb (0.31–0.59); SpOC (0.03–0.27); Pi (1.21–1.94).

Both LOESS and linear regression fitting methods were applied to visualize trends between PVi and co-oximetry parameters. LOESS smoothing ([Fig. 1]) was used to capture local, nonlinear associations that are especially relevant during abrupt perfusion shifts, such as blood loss or transfusion. LOESS plots demonstrated that the PVi–Pi relationship was nonlinear, showing sharp increases during hypovolemic episodes and forming a plateau after fluid resuscitation.

Zoom
Fig. 1 Scatter plots showing the relationship between PVi and (A) SpHb, (B) SpOC, and (C) Pi, with LOESS smoothing curves fitted to each relationship. The LOESS fit illustrates local, nonlinear associations and provides insight into variable behavior during perfusion shifts. All scatter points represent raw, time-aligned intraoperative data. Pi, perfusion index; PVi, pleth variability index; SpHb, spectral hemoglobin; SpOC, spectrophotometric oxygen content.

In contrast, linear regression lines ([Fig. 2]) depict global trends and are overlaid to highlight how variables relate across the entire intraoperative window. All scatter points in both figures represent raw, time-synchronized intraoperative values. The purpose of presenting both plots is to illustrate consistency in relationships, while accounting for the nonlinearity inherent in perfusion dynamics. The PVi–SpHb relationship was weak, confirming that SpHb remained relatively unchanged in the early phases of blood loss.

Zoom
Fig. 2 Scatter plots of PVi versus SpHb, SpOC, and Pi with linear regression trend lines fitted to each. Specific intraoperative events associated with blood loss (annotated in blue) highlight divergence patterns, especially in the Pi versus PVi plot. These linear fits are exploratory and complement the nonlinear analysis in [Figs. 1]. Pi, perfusion index; PVi, pleth variability index; SpHb, spectral hemoglobin; SpOC, spectrophotometric oxygen content.

Multicollinearity was ruled out, with all variance inflation factor (VIF) values below 2.6. Polynomial trend analysis showed a strong nonlinear relationship between PVi and Pi, further emphasizing its role as an early indicator of hemodynamic shifts in this thalassemic patient.


 

Discussion

This correlation-based exploratory analysis highlights the potential of using noninvasive co-oximetry parameters to track and predict perfusion trends during neurosurgery in patients with thalassemia minor. In this pilot case, we did not aim to reproduce PVi by the above model, we only wanted to explore whether trends in Pi, SpHb, and SpOC correlate with PVi in thalassemia, where the perfusion response to volume changes may be altered in case of abnormal hemoglobin. This modeling tests sensitivity overlap, not causality.

In thalassemia minor, the hemoglobin is generally functional in terms of oxygen binding, though there's a quantitative deficit (i.e., slightly lower total hemoglobin levels). Most of the circulating hemoglobin is either HbA (normal adult hemoglobin) or HbA2, which still binds to oxygen normally.[9] SpHb measured via co-oximetry represents total hemoglobin, including both functional and nonfunctional forms (e.g., carboxyhemoglobin, methemoglobin). However, Masimo Radical-7 co-oximeter typically estimates functional hemoglobin (SpHb) unless there is a high level of dyshemoglobins.[10] In this patient, with given stable SpO2 and no evidence of hemolysis or CO exposure, it is safe to assume that the majority of SpHb was functional.

The correction for asynchronous logging of Masimo and Centricity devices in conjunction with harmonizing the data to 1-minute interval resulted in time-matched sampling suitable for exploratory modeling. Pi emerged as the strongest predictor of PVi and it aligned with the physiology of PVi as an indicator of respiratory variation in the plethysmographic waveform that is closely tied to arterial pulsatility and vascular tone. SpHb was relatively stable during mild to moderate blood loss, confirming its lower sensitivity for early hemorrhage detection. SpOC showed modest variation, likely reflecting overall oxygen delivery but without sharp trends. Although in a normal human being these changes are moderate, but in the context of a patient with thalassemia minor, even subtle shifts in parameters like SpHb, Pi, and PVi are physiologically meaningful due to reduced oxygen-carrying capacity. Therefore, we included SpOC and SpHb in the model as their values may signal volume shifts differently in thalassemic patients than the normal patients.

Regression analysis confirmed Pi and SpOC as statistically significant predictors of PVi (p = 0.008 and p = 0.046, respectively), while SpHb had limited impact (p = 0.138), reflecting its slower response to acute hemodynamic shifts. Physiologically, this aligns with expectations. In thalassemia minor, the hemoglobin is typically functional, with most of it being HbA or HbA2. Thus, oxygen-carrying capacity [calculated as: SpHb × 1.34 × SaO2 + (0.003 × PaO2)] was slightly reduced (∼13.8 ml O2/dL) compared with healthy norms, primarily due to mild anemia. The Masimo Radical-7 co-oximeter, under these conditions (normal SpO2, no hemolysis, no CO exposure), reliably reflected functional hemoglobin levels.

The changes in Pi (a marker of arterial pulsatility) and SpOC (a surrogate of oxygen delivery) proved to be more sensitive indicators of volume and perfusion status than SpHb alone. This supports the conceptual use of multivariate models to triangulate early signs of perfusion instability, especially in patients whose baseline hemoglobin is chronically low.

However, this model is limited by its single-patient design, and by the inherent interdependence of physiological variables. Though multicollinearity was statistically ruled out (VIF < 2.6), future validation is essential. Additional limitations include potential overestimation of functional hemoglobin in cases with higher levels of dyshemoglobins (e.g., HbS, HbF, methemoglobin). In such populations, incorporating correction factors from hemoglobin electrophoresis or extended co-oximetry calibration would improve model fidelity.

The clinical utility of this model lies in following situations:

  • Redundancy in monitoring: in high-stakes settings like neurosurgery, waveform-based parameters can intermittently fail. The model allows clinicians to estimate perfusion trends using alternative parameters already displayed on the same monitor (e.g., SpOC, Pi), maintaining situational awareness when PVi is compromised.

  • Insight into pathophysiology in hemoglobinopathies: in thalassemia minor, volume responsiveness may not follow standard trends due to chronically low SpHb and altered oxygen-carrying capacity.[11] In sickle cell disease, there is the issue of chronic anemia and intra-operative oxygenation estimation during neurosurgeries.[12] Likewise, similar concern exists in hemolytic anemia due to autoimmune etiology or underlying enzyme deficiencies.[13] Our model captures how perfusion-related variables interact differently in such patients, which is currently underexplored in the literature.

  • Foundation for predictive analytics: as neuroanesthesia evolves toward real-time analytics and predictive algorithms, exploratory single-case models like ours serve as proof-of-concept. They pave the way for future machine-learning approaches, especially in patients with hematologic disorders.

  • Clinical decision support in monitoring failures: this model can be helpful in settings where Masimo Radical-7 malfunctions due to loss of pleth waveform or power outage. By integrating Pi, SpOC, and SpHb trends, the model provides a fallback method for estimating perfusion dynamics noninvasively. Integration with cerebral oximetry, EtCO2 (End tidal Carbon di-Oxide concentration), PPV (Pulse Pressure Variation), SPV (Systolic Pressure Variation), and stress markers like BIS (BISpectral index) or lactate could further enhance predictive modeling in such cases.

Nevertheless, there are certain limitations. In high levels of dysfunctional hemoglobins, co-oximetry might not always distinguish between oxygen-carrying and nonfunctional hemoglobin.[14] Another caveat is that overestimation of functional oxygen-carrying capacity might occur.[15] In such a scenario, the mathematical model can be augmented by addition of a correction factor for the percentage of nonfunctional hemoglobin (calculated from standard hemoglobin electrophoresis or laboratory calibration). There is still, however, a confounding factor of intraoperative stress conditions (hypotension, hypoxia, blood loss) that affect hemoglobin's oxygen-carrying capacity.[16]

A single case cannot offer statistical generalizability; however, it can still serve as a proof-of-concept for modeling strategies in conditions with altered baseline physiology like hemoglobinopathies. This is just an exploratory modeling for a pilot case study from intraoperative data of a single patient, therefore cross-validation in other patients was not feasible.

However, the modeling framework holds promise for broader application in patients with hemoglobinopathies. Future studies could incorporate corrections for dysfunctional hemoglobin and additional control cases to further validate the relationships and clinical utility.


 

Conclusion

Time-aligned, multi-variable analysis using real-time co-oximetry and perfusion data offers a physiologically interpretable model of PVi behavior during neurosurgery in a thalassemia minor patient. The insights into variable interdependence and real-time trend tracking provide a foundation for noninvasive intraoperative hemodynamic modeling in hematologic disorders.


 

 

Conflict of Interest

None declared.


Address for correspondence

Kunal K. Sharma, DM, MD
Department of Neuroanesthesia Super-speciality Cell, Indira Gandhi Medical College
Ridge Sanjauli Rd, Lakkar Bazar, Shimla 171001, Himachal Pradesh
India   

Publication History

Article published online:
21 August 2025

© 2025. The Author(s). This is an open access article published by Thieme under the terms of the Creative Commons Attribution License, permitting unrestricted use, distribution, and reproduction so long as the original work is properly cited. (https://creativecommons.org/licenses/by/4.0/)

Thieme Medical and Scientific Publishers Pvt. Ltd.
A-12, 2nd Floor, Sector 2, Noida-201301 UP, India


  Permissions and Reprints
Zoom
Fig. 1 Scatter plots showing the relationship between PVi and (A) SpHb, (B) SpOC, and (C) Pi, with LOESS smoothing curves fitted to each relationship. The LOESS fit illustrates local, nonlinear associations and provides insight into variable behavior during perfusion shifts. All scatter points represent raw, time-aligned intraoperative data. Pi, perfusion index; PVi, pleth variability index; SpHb, spectral hemoglobin; SpOC, spectrophotometric oxygen content.
Zoom
Fig. 2 Scatter plots of PVi versus SpHb, SpOC, and Pi with linear regression trend lines fitted to each. Specific intraoperative events associated with blood loss (annotated in blue) highlight divergence patterns, especially in the Pi versus PVi plot. These linear fits are exploratory and complement the nonlinear analysis in [Figs. 1]. Pi, perfusion index; PVi, pleth variability index; SpHb, spectral hemoglobin; SpOC, spectrophotometric oxygen content.