Accuracy and agreement of three blood-loss estimation methods versus the HbMass method for assessing blood loss during PLIF

Abstract

Objective

To evaluate the accuracy and agreement of three widely used blood-loss estimation methods—visual estimation, gravimetric measurement, and the Gross formula method—against the reference HbMass method in patients undergoing posterior lumbar interbody fusion (PLIF).

Methods

A single-center retrospective cohort study included 1000 consecutive elective PLIF patients (2021–2024). Intra-operative blood loss was quantified intra-procedurally by visual, gravimetric, and Gross formula method approaches; HbMass was calculated from pre- and post-operative hemoglobin with patient blood volume estimated by the Nadler equation. Agreement was assessed with Bland–Altman 95% limits of agreement (LoA) and Spearman correlation; sensitivity analyses examined fusion extent, irrigation volume, and sampling timing.

Results

Mean blood loss was 663.8 ± 155.6 mL by HbMass. Visual, gravimetric, and Gross estimates averaged 456.5 ± 175.0 mL, 599.8 ± 167.5 mL, and 608.0 ± 115.0 mL, respectively (all P < 0.001). Correlation with HbMass was negligible (ρ = 0.185), weak (ρ = 0.424), and moderate-to-strong (ρ = 0.742). Bland–Altman biases (95% LoA) were − 238.85 mL (− 631.46, 153.76), − 45.33 mL (− 377.55, 286.90), and − 28.57 mL (− 216.66, 159.52). Sensitivity analyses confirmed robustness.

Conclusion

Among routine methods, the Gross formula method offers the smallest bias and narrowest agreement limits versus HbMass, whereas visual estimation is clinically unreliable. PLIF-enhanced recovery pathways should replace sole reliance on visual assessment with the Gross formula method, supplemented by HbMass in high-risk cases, to optimize peri-operative volume therapy and reduce transfusion-related complications.

Introduction

Perioperative fluid management is a cornerstone of the Enhanced Recovery After Surgery (ERAS) pathway for spinal procedures[1]. Posterior lumbar interbody fusion (PLIF) often results in intraoperative blood loss exceeding 500 mL due to extensive paraspinal muscle dissection, laminectomy, and interbody bone grafting, with a high proportion of hidden blood loss (blood trapped in tissues or surgical sites not captured in suction canisters) [2]. Underestimation may precipitate hypovolaemic hypotension and spinal cord ischaemia–reperfusion injury [3], whereas overestimation can lead to excessive fluid loading, dilutional coagulopathy and unnecessary allogeneic transfusion. Both scenarios are associated with postoperative cognitive dysfunction, increased infection rates and prolonged hospital stay [4]. Accurate quantification of blood loss is therefore prerequisite for optimising volume therapy and improving patient outcome.

Visual estimation, gravimetric measurement and the Gross formula method—currently the most widely used techniques—all have substantial limitations [5]. Visual estimation is highly surgeon-dependent, with reported errors of 30–50% of the true loss. Gravimetric measurement, favoured by anaesthetists and scrub nurses, is more objective but cannot distinguish blood from irrigation fluid, and the weight of fluid retained in swabs is influenced by suction intensity and operative duration. The Gross formula method is based on HbMass balance, yet it assumes a constant circulating blood volume [6], ignoring haemodilution, third-space losses and autologous red-cell salvage, and frequently underestimates actual loss in major spinal surgery.

The HbMass method calculates red-cell loss from the difference between pre-operative and immediate postoperative circulating Hb concentration, using the Nadler equation to estimate patient blood volume (PBV) [7]. It accounts for hidden blood loss and hemodilution effects, making it a validated reference standard.

We retrospectively collected data from 1000 elective PLIF patients. Using Bland–Altman limits of agreement, Spearman correlation and sensitivity analyses, we quantified the systematic bias and random error of visual estimation, gravimetric measurement and the Gross formula method relative to HbMass. These findings may inform selection of an economical yet reliable assessment strategy, transforming “intraoperative blood loss” from an empirical figure into a monitorable ERAS quality indicator.

Materials and methods

Study population

This single-centre retrospective cohort study enrolled consecutive patients who underwent elective single- or multi-level PLIF for lumbar spinal stenosis, spondylolisthesis and/or disc herniation in our orthopaedic department between January 2021 and December 2024.

Inclusion criteria:

1. Definitive diagnosis of lumbar spinal stenosis, spondylolisthesis or disc herniation treated by PLIF;

2. Pre-operative full blood count and coagulation profile obtained within 24 h before surgery;

3. Post-operative full blood count repeated within 48 h;

4. Complete anaesthesia and operation records.

Exclusion criteria:

1. Pre-existing coagulopathy;

2. Female patients menstruating peri-operatively;

3. Pre-operative anaemia (Hb < 110 g/L);

4. Intraoperative transfusion (no autologous re-infusion in this cohort);

5. Multiple-level vertebral fractures, spinal tumor, or infection;

6. Incomplete medical records.

A pilot analysis of 50 patients yielded a blood loss standard deviation of 160 mL. To achieve 90% power (α = 0.05) for detecting a mean difference ≤ 50 mL with 95% LoA width ≤ 350 mL [8, 9], Bland–Altman sample size formula indicated a minimum of 782 evaluable patients. Anticipating a 15% exclusion rate, we initially targeted 920 patients. Continuous enrollment during the study period ultimately yielded 1,000 eligible patients, providing > 90% statistical power.

The study protocol was approved by the Ethics Committee of the General Hospital of Ningxia Medical University (No. KYLL20220133). Written informed consent was waived because of the retrospective design and anonymised data handling. All procedures adhered to the Declaration of Helsinki.

Data collection

Demographics and clinical variables

Age, sex, BMI, ASA class, comorbidities (hypertension, diabetes, coronary artery disease), pre-operative antiplatelet/anticoagulant use, number of fused segments, operative time and surgeon experience (associate vs full professor) were extracted from the hospital information system. Preoperative Hb and Hct were the last values measured before induction; postoperative values were the lowest recorded within 48 h after surgery.

Blood-loss estimation methods

1. Visual

   Before skin closure, the operating surgeon subjectively integrated the suction-bottle scale, area of blood staining in the surgical field, and saturation of surgical pads to assign a single value that was recorded in the operation report as “intraoperative blood loss”.

2. Gravimetric

   Scrub nurses and anaesthetists weighed all dry pads, gauze swabs and the suction canister (containing blood plus irrigation fluid) before incision and again after closure. Net blood weight (g) = (postoperative total weight − preoperative dry weight) − irrigation fluid weight. Volume (mL) = Net weight ÷ 1.050 (whole-blood specific gravity).

3. Gross

   Patient blood volume was first derived with the Nadler equation:

   
   (k₁ = 0.3669, k₂ = 0.03219, k₃ = 0.6041 for both sexes).

   

   where Hct_avg = (Hct_pre + Hct_post)/2 and Hct_post is the lowest postoperative value within 48 h.

4. HbMass

Hemoglobin concentration (g L⁻¹) was measured after induction and within 48 h postoperatively. Red-cell mass loss (g) = PBV (L) × (Hb_pre − Hb_post). Total whole-blood loss = Red-cell mass loss ÷ Hb_pre (no autologous re-infusion in this cohort).

Blinding and quality control

To minimize information bias, two investigators independently extracted raw values in parallel, with discrepancies resolved by a third senior clinician. For statistical analysis, method labels were masked until final model export. All Hb/Hct measurements were performed by a central laboratory on a Sysmex XN-1000 analyzer (two-level QC run twice daily [10]). Data entry used dual-key verification; discrepancies > 1% triggered source verification. Missing data comprised < 1% and were handled by complete-case analysis. Outliers (beyond mean ± 3 SD or clinically implausible values) were reviewed by two senior physicians.

Statistical analysis

SPSS 26.0 and MedCalc 20.0 were used. Normally distributed continuous variables are presented as mean ± SD, non-normally distributed variables as median (P25, P75), and categorical variables as n (%). Blood-loss volumes were compared using repeated-measures ANOVA (following confirmation of normality and sphericity) with Bonferroni-adjusted pairwise tests. Agreement was visualized with Bland–Altman plots and quantified as 95% limits of agreement (LoA). Correlation was assessed with Spearman’s ρ (95% CI) because distributions were non-normal. Sensitivity analyses were performed by: (1) excluding patients with ≥ 3-level fusion; (2) excluding cases with irrigation > 2000 mL; and (3) restricting HbMass postoperative sampling to within 24 h, then recalculating bias, LoA, and ρ. Two-tailed P < 0.05 was considered significant.

Results

Baseline characteristics

A total of 1000 patients undergoing PLIF were enrolled: 542 males (54.2%) and 458 females (45.8%). Mean age was 62.4 ± 10.7 years. Further details are given in Table 1.

Table 1.

General patient data (n = 1,000)

Variable category	Specific item	Value/distribution
Demographics	Gender (male/female)	542 (54.2%)/458 (45.8%)
	Age (years)	62.4 ± 10.7
	BMI (kg/m2)	25.8 ± 3.6
ASA grade	Grade I	171 (17.1%)
	Grade II	721 (72.1%)
	Grade III	108 (10.8%)
Comorbidities	Hypertension	318 (31.8%)
	Diabetes mellitus	189 (18.9%)
	Coronary artery disease	88 (8.8%)
Medication History	Preoperative antiplatelet/anticoagulant use	46 (4.6%)
Surgical factors	Fused segments (1/2/ ≥ 3)	413 (41.3%)/387 (38.7%)/200 (20.0%)
	Operation time (minutes)	135 ± 45
	Surgeon title (associate/full professor)	264 (26.4%) / 736 (73.6%)
Laboratory indicators	Preoperative Hb (g/L)	136.0 ± 13.7
	Postoperative Hb (g/L)	114.5 ± 12.1
	Preoperative Hct (L/L)	0.41 ± 0.05
	Postoperative Hct (L/L)	0.34 ± 0.04

Comparison of estimated intraoperative blood loss

Mean blood loss calculated by the HbMass method was 663.83 ± 155.63 mL. Corresponding values were 456.50 ± 175.00 mL for visual estimation, 599.84 ± 167.45 mL for gravimetric measurement and 608.02 ± 114.96 mL for the Gross formula method. All three methods differed significantly from the HbMass reference (P < 0.001, Table 2), and every pairwise contrast remained significant after Bonferroni correction (P < 0.05, Table 3).

Table 2.

Results of repeated-measures ANOVA

Source	SS	df	MS	F	p-unc	np2
Method	5.26 × 10⁷	3	1.75 × 10⁷	392.7	< 0.001	0.282

Table 3.

Pairwise comparisons (after Bonferroni correction)

Contrast	Adjusted p-value
HbMass vs. Visual	< 0.001
HbMass vs. Gravimetric	< 0.001
HbMass vs. Gross	< 0.001
Visual vs. Gravimetric	< 0.001
Visual vs. Gross	< 0.001
Gravimetric vs. Gross	0.037

Correlation and agreement with the HbMass method

Correlation analysis

Blood-loss data were not normally distributed (Shapiro–Wilk P < 0.001); therefore Spearman’s rank correlation was used. Results are presented in Table 4 and Fig. 1.

Table 4.

Spearman correlation analysis

Method	ρ (95% CI)	Degree of association	p-value
Visual	0.185 (0.12–0.25)	Very weak, clinically negligible	< 0.001
Gravimetric	0.424 (0.36–0.49)	Weak, considerable scatter	< 0.001
Gross	0.742 (0.71–0.77)	Moderately strong, yet some scatter	< 0.001

Fig. 1.

Visualisation of Spearman correlation analyses: A Visual vs. HbMass; B Gravimetric vs. HbMass; C Gross vs. HbMass

Bland–Altman plots (Fig. 2 and Table 5) showed that visual estimation systematically underestimated loss (mean bias − 238 mL) with the widest limits of agreement, indicating poor agreement. Gravimetric and Gross formula method displayed mean biases < 50 mL; the Gross formula method had the narrowest LoA (± ~ 190 mL) and therefore the best agreement with the HbMass reference.

Fig. 2.

Bland–Altman agreement plots: A Visual vs. HbMass; B Gravimetric vs. HbMass; C Gross vs. HbMass

Table 5.

Bland–Altman agreement analysis

Method	Bias (Mean)	SD	LoA_lower	LoA_upper
Visual	-238.85	200.31	-631.46	153.76
Gravimetric	-45.33	169.5	-377.55	286.9
Gross	-28.57	95.96	-216.66	159.52

Sensitivity analyses

To test the robustness of the principal findings, we performed three sensitivity analyses (Table 6).

Table 6.

Summary of sensitivity analyses

Analysis scenario	Cohort size	Method	Bias (mL)	95% LoA (mL)	ρ (95% CI)	Difference from main analysis*
Main analysis	1000	Visual	− 238.9	− 631 to 154	0.185 (0.12–0.25)	–
		Gravimetric	− 45.3	− 378 to 287	0.424 (0.36–0.49)
		Gross	− 28.6	− 217 to 160	0.742 (0.71–0.77)
1. ≥ 3-level fusion excluded	800	Visual	− 232.4	− 620 to 155	0.180 (0.11–0.24)	Bias + 6.5 mL
		Gravimetric	− 40.1	− 365 to 285	0.430 (0.37–0.49)	Bias + 5.2 mL
		Gross	− 25.9	− 210 to 158	0.748 (0.71–0.78)	Bias + 2.7 mL
2. Irrigation > 2,000 mL excluded	902	Gravimetric	− 38.7	− 329 to 252	0.441 (0.38–0.50)	Bias + 6.6 mL
3. Hb sampling within 24 h	812	Gross	− 31.2	− 220 to 157	0.731 (0.70–0.76)	Bias − 2.6 mL

1. Excluding ≥ 3-level fusion (n = 200 excluded; 800 retained)

   Mean biases of the three methods versus the HbMass reference changed by < 10 mL and the widths of the 95% limits of agreement (LoA) varied by < 15 mL, with the direction of bias unchanged, indicating that the number of fused segments has little influence on estimation error.

2. Excluding cases with irrigation > 2,000 mL (n = 98 excluded)

   The mean bias for gravimetric measurement shifted from − 45.3 mL to − 38.7 mL and the LoA narrowed from ± 377 mL to ± 329 mL (change < 15%), suggesting that irrigation volume does not materially affect gravimetric validity.

3. 3. Restricting postoperative Hb sampling to within 24 h (n = 812 retained)

   For the Gross formula method, the mean bias changed from − 28.6 mL to − 31.2 mL, the LoA width varied by < 10 mL and the correlation ρ decreased minimally from 0.742 to 0.731, confirming that the time window for Hb measurement does not appreciably alter the main conclusions.

Discussion

Using contemporaneous data from 1,000 PLIF patients, we compared the accuracy and agreement of four blood-loss estimators. Spearman analysis showed visual estimation correlated very weakly with the reference (ρ = 0.185, 95% CI 0.12–0.25, P < 0.001), making it clinically unreliable. Gravimetric measurement achieved only weak correlation (ρ = 0.424, 95% CI 0.36–0.49) with wide scatter, whereas the Gross formula method reached a moderately strong correlation (ρ = 0.742, 95% CI 0.71–0.77) but still carried substantial individual error. On average, visual estimation underestimated loss by 239 mL with the widest 95% limits of agreement (LoA: − 631 to 154 mL); the gravimetric method underestimated by 45 mL (LoA: − 378 to 287 mL); the Gross formula method underestimated by 29 mL with the narrowest LoA (− 217 to 160 mL). Sensitivity analyses confirmed these findings, indicating that the assessment method itself influences perioperative volume management quality [11].

No internationally accepted gold standard for intra-operative blood loss exists [12, 13]; however, the HbMass method—based on conservation of mass—is recommended as the best available reference [14–16]. It directly quantifies red-cell loss, is theoretically unaffected by infusion fluids, third-space shifts, or irrigation, and requires only readily available preoperative and 48-h Hb values. Its assumption of constant blood volume and omission of salvaged red cells introduce predictable, directionally consistent bias in patients without allogeneic transfusion, making it a dependable comparator in this cohort.

Reported errors of 30–50% for visual estimation align closely with our − 35.9% bias [17]. Previous gravimetric studies in orthopaedics report ρ = 0.30–0.55 [18]; our ρ = 0.424 falls within this range, but earlier work often failed to subtract irrigation fluid, creating spurious precision [19]. In arthroplasty, the Gross formula method can achieve ρ ≈ 0.80 [20], whereas spinal studies seldom exceed 0.60 [21]; our ρ = 0.742 lies between these benchmarks, highlighting that surgical context materially affects formula performance.

With 95% LoA approaching ± 600 mL, visual estimation could mask a grade II haemorrhage (500–1000 mL) in 1 of every 20 patients, delaying fluid or transfusion therapy and increasing risk of hypovolaemic hypotension and spinal cord ischaemia–reperfusion injury [22, 23]. Anaesthetists and surgeons have long lacked a common language for “blood loss”; the bias values reported here can serve as a shared metric for interdepartmental quality dialogue. In high-risk patients (≥ 2 levels, ASA ≥ III), we recommend routinely documenting the Gross formula method value as a second-tier trigger for blood-conservation protocols. Accurate loss estimation underpins goal-directed fluid therapy within ERAS [24]: narrower error margins facilitate earlier recognition of inadequate tissue perfusion, optimise perioperative volume strategy [25], and may reduce lactate-related complications, hospital stay, infection and post-operative cognitive dysfunction.

The present study offers several distinct advantages that enhance its translational value. First, our cohort of 1,000 consecutive patients represents one of the largest head-to-head comparisons of blood-loss estimation methods in spinal surgery, providing robust statistical power and real-world generalizability. Our findings demonstrate consistent performance metrics across diverse patient phenotypes, including multi-level fusion procedures and varying ASA classifications. Second, the Gross formula method has minimal equipment requirements—requiring only a calculator and routine hemoglobin values—rendering it highly cost-effective compared with isotope-based assays or advanced hematology tests. This practicality allows for immediate implementation in most hospitals, particularly in resource-limited settings. Third, the methodological framework is readily adaptable to other spinal procedures and broader orthopedic surgeries. Although prospective validation remains essential, the immediate clinical utility and scalability of our approach represent a meaningful step toward precision perioperative care.

Limitations

Our single-center retrospective design reflects institutional surgical routines, haemostatic preferences, and laboratory platforms; multicenter validation is warranted. Although “retrospective blinding” was applied, intraoperative visual and gravimetric values were recorded by clinicians and thus remain subject to observer bias. HbMass, though the best internal reference, is not an absolute gold standard; it ignores red-cell trafficking, iron kinetics and volume fluctuations. Future studies should verify findings using isotope-labelled or carbon-monoxide haemoglobin methods. We did not capture post-operative drain losses or hidden blood loss beyond 72 h, potentially underestimating total bleeding, and we did not stratify for different topical haemostatic agents or anti-fibrinolytics—factors that might confound results [26, 27].

Conclusion

Against HbMass reference, visual, gravimetric and Gross formula method estimates differed significantly during PLIF. Visual estimation produced the largest error (− 239 mL bias, 95% LoA − 631 to 154 mL), gravimetric measurement yielded moderate error, and the Gross formula method demonstrated the smallest bias (− 29 mL) and strongest correlation (ρ = 0.742). While surgeons favor visual assessment and anesthetists prefer gravimetric data, perioperative volume management should reduce sole reliance on visual estimation and adopt the Gross formula method as the preferred alternative. In high-risk patients (≥ 2-level fusion, ASA ≥ III), concurrent HbMass calculation provides a quality-control metric within ERAS pathways to guide fluid and transfusion decisions, mitigating risks of both hypovolemia and excessive infusion, and ultimately improving patient outcomes.

Author contributions

QPM was responsible for drafting the primary manuscript. XYD was responsible for data collection. XWW and JL contributed to the refinement, and conducted a review for grammatical and spelling errors. NKN provided guidance on the innovation of the writing, and ensured the scientific accuracy and rigor of the paper.

Funding

This project is financially supported by The First Batch of Dominant Discipline Cluster Construction Projects of General Hospital of Ningxia Medical University (No. YSXKQ2024004) and Ningxia Hui Autonomous Region Health Scientific Research Key Project (2025-NWZD-A001).

Data availability

If you require the raw data, please contact the corresponding author at niuningkui6743242@163.com.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Review Committee of the General Hospital of Ningxia Medical University (No. KYLL20220133). All procedures were conducted in accordance with the ethical standards established by the Declaration of Helsinki and its subsequent amendments.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.