Utility of pre‐operative haemoglobin concentration to guide peri‐operative blood tests for hip and knee arthroplasty: A decision curve analysis

Abstract Objective Assess the prognostic value of pre‐operative haemoglobin concentration (Hb) for identifying patients who develop severe post‐operative anaemia or require blood transfusion following primary total hip or knee, or unicompartmental knee arthroplasty (THA, TKA, UKA). Background Pre‐operative group and save (G&S), and post‐operative Hb measurement may be unnecessary for many patients undergoing hip and knee arthroplasty provided individuals at greatest risk of severe post‐operative anaemia can be identified. Methods and Materials Patients undergoing THA, TKA, or UKA between 2011 and 2018 were included. Outcomes were post‐operative Hb below 70 and 80 g/L, and peri‐operative blood transfusion. Logistic regression assessed the association between pre‐operative Hb and each outcome. Decision curve analysis compared strategies for selecting patients for G&S and post‐operative Hb measurement. Results 10 015 THA, TKA and UKA procedures were performed in 8582 patients. The incidence of blood transfusion (4.5%) decreased during the study. Using procedure specific Hb thresholds to select patients for pre‐operative G&S and post‐operative Hb testing had a greater net benefit than selecting all patients, no patients, or patients with pre‐operative anaemia. Conclusions Pre‐operative G&S and post‐operative Hb measurement may not be indicated for UKA or TKA when adopting restrictive transfusion thresholds, provided clinicians accept a 0.1% risk of patients developing severe undiagnosed post‐operative anaemia (Hb < 70 g/L). The decision to perform these blood tests for THA patients should be based on local institutional data and selection of acceptable risk thresholds.


| INTRODUCTION
The number of elective primary hip and knee arthroplasties performed internationally is projected to increase exponentially for hip arthroplasty by 2050 and four-fold for knees by 2030, with comparable projections across 20 Organisation for Economic Cooperation and Development countries. [1][2][3] Hip and knee arthroplasties are highly successful procedures at improving quality of life, but the surgery can result in significant blood loss. 4,5 Blood loss can exceed 1 L during the peri-operative period for a primary hip or knee arthroplasty procedures when accounting for visible and hidden blood losses. 6 Despite blood conservation strategies, over 80% of patients are anaemic on discharge from hospital and up to 5% require allogenic red blood cell transfusion. [7][8][9][10] The recommended haemoglobin concentration (Hb) threshold for administering allogeneic red blood post-operatively is 70 g/L, or 80 g/L in the presence of acute coronary syndrome. 11 Routine clinical practice is to perform group and save (G&S) in all patients prior to arthroplasty surgery and a post-operative blood test for Hb prior to discharge from hospital, however, there is limited guidance for when these investigations are indicated or can be omitted.
Pre-operative G&S and post-operative Hb measurement may be unnecessary when the risk of requiring a blood transfusion or a patient developing severe post-operative anaemia is acceptably low.
Previous studies have identified pre-operative Hb thresholds that predict blood transfusion using Receiver Operator Characteristics (ROC) curve methodology, choosing pre-operative Hb values associated with the highest combined sensitivity and specificity, and giving equal weighting to true and false positives. 12,13 However, this does not allow clinicians to consider the harm: benefit ratio they feel is appropriate for their patient cohort.
The potential benefits of a targeted approach to peri-operative blood tests include reduced cost, improved patient experience, and shorter length of inpatient stay by preventing delays waiting for blood results and facilitating safe day-case surgery. Potential risks of omitting these investigations include having no blood product available when required and discharging patients with severe undiagnosed anaemia. The acceptable harm: benefit ratio will differ between institutions and clinicians.
The aim of this study was to characterise the prognostic value of pre-operative Hb for identifying patients who develop severe postoperative anaemia or require blood transfusion following elective primary hip or knee arthroplasty. We address four research questions:   (Table S1).
THA were performed using the anterolateral or posterior approach using cemented and uncemented implants. TKA were mostly performed using the medial parapatellar approach and a tourniquet. UKA were routinely performed with a minimally invasive medial arthrotomy and a tourniquet. Routine tranexamic acid administration was introduced after 2015 as intravenous with or without topical administration. Patients who had a blood transfusion within 3 months prior to surgery were excluded as they are likely to have an underlying medical condition that requires regular Hb measurement. Procedures with a missing date and duplicate entries on electronic health records were also excluded.

| Outcomes
The primary outcome was the National Institute for Health and Care Excellence (NICE) recommended transfusion threshold of a Hb less than 70 g/L within 7 days following surgery (<70 g/L transfusion threshold: yes/no, binary). 11 Secondary outcomes were the NICE recommended transfusion threshold for patients with acute coronary syndrome of a Hb less than 80 g/L (<80 g/L transfusion threshold: yes/no, binary) 11 ; and administration of a blood transfusion intra-or post-operatively (blood transfusion: yes/no, binary) within 7 days following surgery.
All data collected between 2011 and 2018 were used for Hb measurements. Data collected between 2013 and 2018 were used for the blood transfusion analysis when electronic prescribing for blood transfusion was introduced.

| Variables
Data were extracted for: age (years, continuous); sex (male/female, binary); pre-operative Hb (g/L, continuous); American Society of Anesthesiologists (ASA) classification (range: 1-5, ordinal); and postoperative Hb (g/L, continuous). The closest pre-operative Hb value to surgery start date and time was used from within 6 months prior to surgery, and the closest value to surgery end date and time was used from within 7 days following surgery.
Pre-operative and post-operative Hb were categorised according to the World Health Organisation (WHO) and the International Consensus Statement (ICS) definitions of anaemia for descriptive purposes. 16,17 The WHO defines anaemia using a haemoglobin <120 g/L for females and <130 g/L haemoglobin for males; and the ICS defines anaemia using a <130 g/L for both females and males.

| Regression analysis
The association between pre-operative Hb and the outcomes was evaluated using unadjusted and adjusted logistic regression. The variables age, sex, and ASA classification were chosen a-priori to be included in the adjusted logistic regression. We assessed non-linearity and checked the functional form of pre-operative Hb using fractional polynomials of degree 1. 18 Unadjusted and adjusted odds ratios with 95% confidence intervals are presented.

| Decision curve analysis
Decision curve analysis (DCA) was used to assess and compare the utility of four intervention strategies to identify patients with postoperative anaemia and needing a blood transfusion, and therefore, which patients should be selected for G&S prior to surgery and similarly, which patients should have post-operative blood tests to measure their Hb. 19,20 Intervention strategies were:

Risk thresholds-Clinician's preference
We considered a range of 0.1% (clinician is more concerned about undiagnosed post-operative anaemia) to 1% (clinician is more concerned about unnecessary post-operative blood tests) that a clinician may find acceptable for not having performed a G&S when a postoperative blood transfusion is indicated. As the risk threshold decreases there will be more true positives (necessary G&S and postoperative blood tests) at the expense of more false positives (unnecessary G&S and post-operative blood tests). When a risk threshold of 0.1% (1:1000) is used, we perform a G&S and post-operative blood test on at most 1000 patients per one true positive.
Alternatively, the odds of the risk threshold represent the maximum number of false positives that a clinician is willing to accept per true positive; known as the 'harm to benefit ratio'. For the 0.1% threshold, the odds (harm to benefit ratio) are 1:999, hence at most 999 false positives are accepted per one true positive. Therefore, risk thresholds of 0.25%, 0.5%, 0.75% and 1% imply a clinician is willing to unnecessarily perform G&S and a post-operative blood test on 399, 199, 133 and 99 patients to correctly identify one patient where a transfusion is required, respectively.
As the risk threshold decreases, the clinician is more averse to the risk of not having blood available for transfusion or not diagnosing severe post-operative anaemia. The range of risk thresholds refers to the clinician's preference to perform a G&S and post-operative blood test for a given patient, which can be discussed with the patient and thus can vary accordingly.

Net benefit
To assess the utility of the four intervention strategies for selecting patients for G&S and post-operative Hb measurement we calculated net benefit, which is the net proportion of true positives, much like net profit equals revenue minus all expenditures in business.
The benefit of an intervention strategy is that is correctly identifies which patients had a post-operative Hb < 70 g/L, <80 g/L or had a blood transfusion (and hence needed a G&S and postoperative blood test). For example, a net benefit of 0.002 is equivalent to having 2 additional patients correctly identified as needing blood transfusion per 1000 patients, without incorrectly selecting anyone who did not need a blood transfusion.
The net benefit is directly comparable to the 'test none' strategy (do not perform pre-operative G&S or post-operative Hb measurement for anyone) which by definition has a net benefit of 0 because if you never intervene you do not have any true or false positives. The difference in net benefit between all strategies can also be calculated, for example, if the net benefit increases from 0.001 when using 'Test all: intervention for all patients' to 0.002 when using the 'Test patients according to risk of post-operative anaemia' strategy, the latter has 1 more net detected blood transfusion per 1000 patients for the same number of unnecessary G&S and post-operative blood test carried out.
The net benefit was calculated and presented in a decision curve at risk thresholds between 0.1% and 1%, as it was deemed unlikely a clinician would risk not having blood available when the risk of needing it exceeds 1%. The DCA was informed using an unadjusted logistic regression analysis of pre-operative Hb for the intervention strategy based on institution and procedure specific pre-operative Hb. It is presented for each outcome (Hb < 70 g/L, Hb < 80 g/L, allogeneic blood transfusion) and each risk threshold (0.1%, 0.25%, 0.5%, 0.75%, and 1%).

Net reduction
We also calculated the test trade-off, which is the net reduction in the number of unnecessary G&S and post-operative blood tests (false positives) using procedure and institution specific pre-operative Hb F I G U R E 1 Flow of patients and procedures into the study. NOC, Nuffield Orthopaedic Centre, Oxford, UK; THA, total hip arthroplasty; TKA, total knee arthroplasty; UKA, unicompartmental (partial) knee arthroplasty; DCA, Decision curve analysis. *OPCS codes used to identify THA, TKA and UKA procedures: W371, W381, W391, W941, W421, W401, W581 cut-offs compared to the 'Test all: intervention for all patients' strategy. This number is equivalent to the number of avoided interventions, whilst keeping the number of true positives the same. For example, a net reduction of 0.04 means that, per 100 patients, four unnecessary G&S and post-operative blood test interventions are avoided for the same level of necessary interventions. We also calculated and plotted the net reduction, over the risk thresholds.
Further details about the net benefit and net reduction calculations are provided in Supporting Information S1: Appendix A. The utility of testing patients according to risk of post-operative anaemia was compared with all other intervention strategies for each outcome and procedure type.

| Additional analyses
We performed a sensitivity analysis using data from years 2015 to 2018 to account for any potential temporal changes in patient blood management and to reflect current practice. We also performed a sensitivity analysis for transfusions that took place prior to a formal postoperative Hb measurement which assumed that these patients had Hb < 70 g/L and Hb < 80 g/L before their transfusion. Missing data were described, and no imputation analysis was performed given the small amount of missing data. All analyses are complete case analyses.

| RESULTS
44 612 surgical procedures were undertaken at the NOC between January 2011 and August 2018. Of these, 10 015 procedures were THA, TKA and UKA procedures (Figure 1).     3.2 | What are the temporal trends of postoperative Hb below the 70 and 80 g/L transfusion thresholds, and allogeneic blood transfusion following elective primary hip and knee arthroplasty?

| Baseline and post-operative characteristics
The proportion of patients receiving blood transfusions fell during the period of study, although the proportion of patients developing postoperative anaemia remained stable (Figure 2).

|
What is the relationship between preoperative Hb, and post-operative Hb and blood transfusion?
Odds of patients developing a post-operative Hb below 70 or 80 g/L, and patients receiving an allogeneic blood transfusion reduced per unit increase in pre-operative Hb for THA and TKA procedures ( Figure S1 and 3.4 | What are the pre-operative Hb thresholds for different risks of severe post-operative anaemia? Pre-operative Hb thresholds for different risks of post-operative anaemia differed between THA and TKA (Table 3). An example scenario based on these results is a clinician who does not wish to unnecessarily collect blood for pre-operative G&S in more than 99 patients to identify 1 patient with a post-operative Hb less than the 80 g/L transfusion threshold (1% risk threshold). They would select a pre-operative Hb < 136 g/L to decide whether to perform a pre-operative G&S for THA. In our cohort, a threshold of <136 g/L would have meant 130 patients who subsequently required a blood transfusion had a pre-operative G&S performed preoperatively, and eight patients who subsequently required a blood transfusion did not have a G&S performed preoperatively and would need one prior to transfusion. G&S from 2478 patients would have been collected but not required, and no G&S would have been collected from 2177 patients who did not require a blood transfusion.
3.5 | What is the utility of different strategies using pre-operative Hb to select patients for preoperative G&S and post-operative Hb measurement?
Decision curve analysis showed that for THA and TKA, using the for transfusion threshold analyses, and 2013-2018 for the blood transfusion analysis. Analysis was not performed for a 70 g/L transfusion threshold for TKR due to the paucity of event rates. The y-axis is the benefit (net benefit), and the x-axis is the preference (threshold probability). The benefit of an intervention strategy is that is correctly identifies which patients had a post-operative Hb < 70 g/ L, <80 g/L or had a blood transfusion (and hence actually needed a G&S and postoperative blood test). The preference refers to how clinicians value different outcomes and what risk they are willing to take to not have blood when it is needed for a given patient, which varies. For example, if Clinician A was concerned about undiagnosed post-operative anaemia in their THA (Hb < 70 g/L) patient, they may only be willing to accept a 0.1% risk to not have blood available when it is needed, compared to Clinician B who is more concerned about unnecessary blood test and would be willing to accept a slightly higher 1% risk of not having blood when it is needed. In this case for Clinician A, there is little difference between the intervention strategies of performing a G&S and post-operative blood test on all patients, only on patients with pre-operative anaemia, or on patients at risk of post-operative anaemia. However, Clinician B would be better to use the intervention strategies of performing a G&S and post-operative blood test only on patients at risk of post-operative anaemia based on pre-operative Hb measurement.
compared with the 'Test none' and 'Test patients with preoperative anaemia' strategies ( Figure 3). A net reduction in unnecessary blood tests was found for these outcomes when using the 'Test patients according to risk of post-operative anaemia' strategy, compared to the 'Test all' strategy ( Figure S2 and Additional data of net benefit and net reduction comparisons between intervention strategies are provided in Figure S2 and Table S3. Only two patients undergoing TKR developed a post-operative Hb below 70 g/L, and one patient undergoing UKR developed a post-operative Hb below 80 g/L, hence decision curve analysis was not performed for these outcomes.  (Figures S3 and S4, Table S4).

| Sensitivity analysis
Thresholds for intervention and the associated net benefit for using Hb thresholds based on the 'Test patients according to risk of post-operative anaemia' strategy increased when THR patients who received a blood transfusion before their postoperative Hb was T A B L E 4 Risk associated pre-operative Hb thresholds and decision curve analysis results for THA and TKA using data limited to 2015-2018. measured were assumed to have a Hb < 70 g/L. A smaller increase was found in the thresholds for intervention and associated net benefit when both THR and TKR patients, who received a blood transfusion before their postoperative Hb was measured were assumed to have a Hb < 80 g/L. Full results are provided in Table S5.

| Literature
Routinely performing blood tests for pre-operative G&S and postoperative Hb measurement is potentially unnecessary given the low prevalence of severe post-operative anaemia and blood transfusion. 21 When unnecessary, these blood tests increase cost and impair patient experience through venepuncture. They may also increase preoperative hospital attendance delay surgery (awaiting G&S), or discharge from hospital (awaiting post-operative Hb). Previous studies with smaller sample sizes proposed thresholds for performing these investigations at 130 g/L or between 121 and 124 g/L depending on age. 12,22 The ROC analysis used in these studies have limited clinical interpretability and assumes true and false positives are equally important, which may not be accurate. When selecting patients for G&S, a true positive potentially has a higher misclassification cost and carry greater importance than a false positive. 23 The ability to stratify risk using decision curve analysis, enables the selection of a threshold for G&S and post-operative Hb measurement that is deemed appropriate for a specific clinical setting. These thresholds can be used for electronic clinical decision support tools increasingly adopted for preoperative assessment. In addition to the patient population and clinical practice, institutional factors may influence selection of appropriate harm to benefit ratios, such as the availability of electronic blood issue and proximity to blood banks.
Clinicians must also assess the risk discharging patients with severe undiagnosed anaemia, and a different harm to benefit ratio Hb threshold may be used to select patients for G&S than for post-operative

| Strengths and limitations
Our study uses contemporary and clinically relevant statistical methodology to evaluate the utility of different strategies to select patients for pre-operative G&S and post-operative Hb measurement. We build on existing ROC curve analyses that assume true and false positives are equally important 24,25 and use decision curve analysis which better addresses the utility of a prognostic test. Decision curve analysis weights true and false positives separately, and does not require categorisation of pre-operative Hb to derive clinically informed risk thresholds, which minimises the loss of statistical information. 26  based on achieving a conservative 15% of the maximum r-squared and a 0.47% event rate (lowest event rate was found for the 70 g/L outcome). 27 A minimum of 457 patients with six events (events per predictors = 5.80) is required for the TKA analyses, based on achieving a conservative 15% of the maximum r-squared using one predictor and a 1.27% event rate (lowest event rate was found for the 80 g/L outcome). The heuristic shrinkage for each analysis ranged from 0.97 to 0.99, indicating likely very small amounts of optimism in our estimates. Missing data was described in our study but was not imputed.
Given the very small amount of missing data present, a complete case analysis was justified. 28 We used an unadjusted logistic regression model to inform the pre-operative Hb thresholds for different risk of

| Future research
This study was performed on data from a single institution, but the methodology can be applied to different surgical procedures and different institutions. Clinicians overseeing the care of patients receiving primary hip and knee arthroplasty may wish to select an acceptable harm to benefit ratio Hb threshold from our study and then retrospectively validate the performance using their local data. There may be significant financial savings associated with reducing the number of unnecessary G&S and post-operative blood tests, and a formal health economic review may be of value in the future.

| Conclusions
In summary, our study outlines a means of selecting patients for G&S and post-operative Hb measurement based on the risk of developing severe post-operative anaemia. Pre-operative G&S and post-operative Hb measurement may not be indicated for UKA or TKA when adopting restrictive transfusion thresholds provided clinicians accept a 0.1% risk that patients will develop severe undiagnosed postoperative anaemia (Hb < 70 g/L). The decision to perform these blood tests for THA patients should be based on local institutional data and selection of an acceptable harm to benefit ratio. Decision curve analysis can be adopted for all surgical procedures to accurately select patients who require pre-operative G&S and post-operative Hb measurement, facilitating day-case surgery and reduced cost.