#### What We Already Know about This Topic

#### What This Article Tells Us That Is New

^{1}Nevertheless, the principal factor limiting reduction in testing is insufficient evidence regarding indications.

^{1}In this paper, we consider the decision to perform ABO and Rh(D) blood typing and identification of clinically important red cell antibodies (“type and screen”) before elective procedures.

*e.g.*, 0.8%) and relatively high incidences of type and screen (

*e.g.*, 49%).

*e.g.*, median) EBL. Friedman created the MSBOS using many hospitals' data, suggesting type and screen for procedures with “minimal blood loss.”

^{2}Richardson

*et al.*recommended type and screen, not cross match, for patients undergoing cholecystectomy because “the expectation of blood loss was low.”

^{3}Another study objective was to determine how to apply prior knowledge from the median EBL of the procedure when making type and screen recommendations for a procedure. We hypothesized that among procedures with minimal EBL, the

*a priori*probability of transfusion is low, and additional information on the observed incidence of transfusion for the procedure can be used to revise the probability.

^{4}

^{–}

^{6}Jayaranee

*et al.*recommended type and screen for procedures with a mean more than 0.3 units per case.

^{4}Mahadevan suggested type and screen for a mean of more than 0.5 units per case.

^{6}Cheng

*et al.*recommended that the MSBOS be updated using data from the anesthesia information management system, with type and screen performed for procedures having a mean of more than 0.5 units per case.

^{6}Using data from the anesthesia information management system, the facility's MSBOS can be revised automatically for all procedures, not just updated manually for common procedures. In this study, we used our anesthesia information management systems data for this purpose, and hypothesized that thresholds of mean of 0.3–0.5 units per case would be valid statistically for some procedures.

^{7}

^{–}

^{9}van Klei

*et al.*showed that not using type and screen preoperatively for procedures with the rate of transfusion less than 5% was suitable for cholecystectomy at multiple facilities.

^{8},

^{9}We therefore examined the relationship between this standard of “5% probability of transfusion” and the median EBL of procedures. We hypothesized that for most procedures with relatively small sample sizes (

*e.g.*, fewer than 163 cases)**,

^{10}

^{–}

^{16}the historical estimated median EBL could be used to forecast whether the risk of transfusion is less than the 5% criterion. Finally, we compared the incidence of type and screen in our study with that using current practice.

#### Materials and Methods

*All*use of the term “procedure” in the paper refers to the scheduled procedure, since the decision to type and screen is made in reference to scheduled procedures, not actual (performed) procedures.

^{15},

^{17}No statistical power analysis was performed

*a priori*because the objective initially was to describe the functional relationships described in the Introduction (

*i.e.*, there was insufficient knowledge to perform the calculations).

*e.g.*, more than 100 units administered for EBL less than 20 ml). Of the 21 cases, five had procedures of median EBL less than 50 ml, 0.004% of such records. The final data set contained 160,207 cases in adults of 1,253 elective procedures (table 1).

^{15},

^{18},

^{19}

##### Statistics

^{4}

^{–}

^{6}The sample mean is the maximum likelihood estimator for the Poisson distribution's single parameter, the mean. Equality of the mean and variance is a mathematical property of the Poisson distribution. This is in contrast, for example, to a binomial distribution, where the mean and the variance are different parameters. Equality of the mean and variance of erythrocyte units transfused was tested for each procedure by comparing (n − 1) × (sample variance)/(sample mean) to a chi-square distribution with (n − 1) degrees of freedom.

^{20}

*e.g.*, incidence of type and screen) and percentiles (

*e.g.*, EBL) were calculated using the Clopper-Pearson method.

^{21},

^{22}The computational accuracy of the relevant equation (table 2) in Office Excel 2010 (Microsoft, Redmond, WA) was better than within 0.001%.††

#### Results

##### Criteria for the Type and Screen Decision

*et al.*previously suggested that type and screen be performed for procedures with 5% or more incidence of transfusion with one or more units.

^{8},

^{9}They showed that the criterion was suitable for multiple facilities, but studied only a few procedures, all with large numbers of cases.

^{8},

^{9}

*et al.*

^{8},

^{9}

*e.g.*, 5.0% or more) differs from one based on the mean units transfused for each procedure (

*i.e.*, the “transfusion index”).

^{4}

^{–}

^{6},

^{23},

^{24}One of our objectives was to interpret use of the mean units transfused. We start by considering the procedure of unilateral total knee replacement. The procedure had an incidence of type and screen of 98.3%. There was transfusion of 0 units for n = 2,417 cases, 1 unit for 262 cases, 2 units for 4 cases, and 3 units for 1 case. The mean equals 0.102 units, where 0.102 = (0 × 2,417 + 1 × 262 + 2 × 4 + 3 × 1)/(2,417 + 262 + 4 + 1). This mean would be related to the incidence of transfusion provided the number of units transfused per case follows a Poisson distribution§§.

^{25}Setting the probability equal to 5.0% and solving for the unknown mean gives 0.051. Thus, if the mean number of units transfused exceeds 0.051 units, each patient would have a probability of transfusion exceeding 5.0%. The mean is slightly larger than the corresponding probabilities (

*e.g.*, 0.051 more than 5.0%) because occasionally more than one unit is transfused. Still, the issue of 0.051

*versus*5.0% (

*i.e.*, 0.050) is minor considering that Mahadevan and Cheng

*et al.*recommended 0.5 units as the threshold to perform a type and screen (

*i.e.*, a value effectively 10 times larger than the 5.0% threshold).

^{5},

^{6}We address this discrepancy below in the section Scheduled Procedures with Minimal Median EBL (Less Than 50 ml) and in the appendix.

*i.e.*, the “transfusion index”) were equivalent, we evaluated fits to Poisson distributions.## For some procedures, the Poisson distribution appeared reasonable. For example, above we reported data from the study hospital for total knee replacement. The sample mean transfusion was 0.102 units, very close to the sample variance of 0.097 units, confirming a Poisson distribution (

*P*= 0.97). In contrast, for some other procedures, the sample means and variances were markedly different (

*i.e.*, poor fit to a Poisson distribution). For example, consider erythrocyte transfusions for the procedure of combined anterior and posterior cervical fusion. The procedure had an incidence of type and screen of 97.3%. Among the 541 cases, there were 500 with 0 units, 24 with 1 unit, eight with 2 units, four with 3 units, three with 4 units, one with 7 units, and one with 12 units. The sample mean equaled 0.153 units whereas the sample variance equaled 0.593 units,

*P*< 0.0001. Overall, there were 317 procedures with sample mean of more than 0 units, excluding procedures with just n = 1. For 61% of these 317 procedures, the data were inconsistent with Poisson at

*P*< 0.05. This incidence was considerably greater than the 5% incidence expected at random if a Poisson distribution were a reasonable fit for many procedures. Similarly, for 56% of the procedures, the data were inconsistent with Poisson at

*P*< 0.01, rather than the 1% incidence expected at random. Thus, recommendations based on the mean transfusion and assuming a Poisson distribution would be unreliable. Instead, we rely on a 5.0% incidence of transfusion as the threshold for type and screen, not on a specific probability distribution and its parameters.

##### Classification of Scheduled Procedure and Missing Values for EBL

*i.e.*, to make the decision whether to type and screen), the procedure needs to be specified. The hospital had 1,253 procedures using its internal classification system, many of which were performed rarely. For procedures that have less than 19 cases, the fifth percentile of the observed data cannot be calculated. There were 2.5% of cases for which there were less than 19 cases per procedure (95% confidence limit less than 2.6% of cases).‖‖

^{26}The larger problem is that to judge reliably when incidences differ from 5.0% based solely on observed numbers of cases with and without transfusion, sample sizes less than 163 are small.** For 18.1% of cases (95% confidence limit less than 18.2%) there were less than 163 cases per procedure.

*via*a crosswalk process, facilitated by use of the anesthesia CPT code for the performed (not scheduled) procedure from the department's billing office. The use of anesthesia CPT did result in fewer cases unable to be analyzed because of sample size considerations. For example, the anesthesia CPT 0750 encompassed six of the studied hospital's surgical procedures, including umbilical hernia repair (n = 426, median EBL less than 10 ml) and Spigelian hernia repair (n = 7, median EBL less than 10 ml). None of the 488 cases with anesthesia CPT 0750 included transfusion (

*i.e.*, the anesthesia CPT predicted no transfusion).

^{27},

^{28}For example, among EBL near the 50 ml threshold used in the section Scheduled Procedures with Minimal Median EBL (Less Than 50 ml), there were 7% overestimation for EBL 25 ml, 23% overestimation for EBL 30 ml, 10% underestimation for EBL 60 ml, and 11% underestimation for EBL 70 ml.

^{27},

^{28}Multiplying, each of these biases was less than 10 ml, negligible relative to the 500 ml range of median EBL in figure 1. The experimentally measured bias could substantively influence the created MSBOS for EBL near the threshold of 50 ml. However, actual reported EBL have digit bias. Among recorded EBL of 40–60 ml, 99.3% are 40 ml, 50 ml, or 60 ml (n = 12,201/12,292).

*post hoc*. We rely only on scheduled procedures, which frequently

^{15}differ from actual procedures.

*e.g.*, 10 ml) that could not possibly have been measured, the anesthesia provider made no entry. In contrast, for total hip revision with a 36% incidence of transfusion (n = 298/822), the EBL was missing for only 1.6% of cases. This explanation was tested quantitatively, among procedures with 19 or more cases, by treating procedure as a fixed effect in logistic regression. The presence (Y) or absence (N) of a listed EBL significantly increased the incidence (Y

*vs.*N) of transfusion (

*P*< 0.0001, odds ratio 1.45, 95% CI 1.30–1.67). When the analysis was repeated using a larger threshold, specifically 59 or more cases, the results were effectively the same (

*P*< 0.0001, odds ratio 1.45, 95% CI 1.27–1.63). Thus, the anesthesia providers were more likely to enter an EBL when there was transfusion. Imputing the EBL for cases for which the EBL was unstated by using EBL for those cases of the procedure with an EBL listed would result in bias in estimates of the relationship between EBL and transfusion.

*e.g.*, 1.6% of total hip revision cases in the preceding paragraph). Figure 1 shows transfusion plotted against EBL without stratification by procedure. The corresponding odds ratio between presence (Y) or absence (N) of a listed EBL and transfusion (Y) was 17.2 (95% CI 15.5–19.1). The median EBL was used, as compared with other statistics such as the mean or mode, so that substitution of any value for the few absent EBL causes no change in the summary EBL for such procedures. For example, for total hip revision, from results of the preceding paragraph, absent EBL likely were less than the median EBL of 500 ml. Nevertheless, the calculated median EBL would be the same whether all absent values were treated as equaling 10 l, 10 ml, or the median EBL itself of 500 ml.

*does not imply*that there was no blood loss, just the EBL of the cases with absent EBL were not larger than the median EBL of the procedure (see Discussion).

##### Scheduled Procedures with Minimal Median EBL (Less Than 50 ml)

^{29}Among cases of procedures with most EBL absent or equaling 0 ml, the hemoglobin was checked preoperatively for 37% of cases (n = 34,217/92,100). Among cases for which the procedure had median EBL larger than 0 ml but less than 50 ml, the hemoglobin was checked preoperatively for 84% of cases (n = 22,318/26,492). Each further increase in the median EBL was associated with an increase in the percentage of cases with the hemoglobin checked preoperatively. For example, there were 180 procedures with at least 19 or more cases per procedure and median EBL larger than or equal to 50 ml and less than or equal to 500 ml. Applying linear logistic regression to these 39,259 cases, procedures with a median EBL of 50 ml and 500 ml had overall estimated incidences of hemoglobin checked preoperatively of 93% and 96%, respectively (

*P*< 0.0001).

*e.g.*, n = 20), unless the observed proportion transfused was significantly different from 0.3%, the Bayesian estimated incidence of transfusion was close to 0.3%. For example, among the 20 patients who underwent endometrial ablation with balloon, none was transfused. The Bayesian estimate was not 0.0% but closer to 0.3%. The Bayesian random effect effects analysis implied shrinkage of the procedure's observed incidence of transfusion toward the underlying incidence of 0.3% among all procedures.

*via*the Clopper-Pearson method,

^{21},

^{22}using the observed number of cases with and without transfusion for the procedure. Type and screen would be performed for those procedures for which the lower limit exceeds 5.0%. The lower limit applied because,

*a priori*, the probability of transfusion was low based on choosing only procedures with minimal median EBL. A large incidence of transfusion would need to be observed for a procedure to warrant type and screen.

^{21},

^{22}:

*P*< 0.0001). The sample size was smaller (n = 87) for assessing incidences of type and screen than transfusion (n = 171) because patients who had undergone surgery during the prior 30 days could not be studied, as their type and screen may have been for the previous procedure(s).

*i.e.*, the “transfusion index”) of more than 0.50 units,

^{5},

^{6}even though by Poisson distribution this would correspond to a prediction that 50% of patients would have been transfused. Both of the procedures with a lower 95% confidence limit for incidence of transfusion more than 0.05 had means of more than 0.50 units. Thus, recommendations matched, because the Poisson distribution assumption did not hold. The recommendations are, in contrast, strikingly inconsistent for larger median EBL (see appendix).

*P*< 0.0001), but the reduction was substantively minor (1.6%). The statistical significance among procedures with larger EBL further suggests validity of the preceding results.

#### Discussion

^{5},

^{6}and for procedures with

*P*> 5.0%

^{8},

^{9}for erythrocyte transfusion. From sections Criteria for the Type and Screen Decision and Scheduled Procedures with Minimal Median EBL (Less Than 50 ml) and appendix, these two previously developed and applied criteria differ 10-fold and result in divergent recommendations. For example, if the 0.50 unit (transfusion index) criterion was used, then patients scheduled to undergo total knee replacement or anterior and posterior cervical fusion would not undergo type and screen, even though the current incidence exceeds 97%.

*e.g.*, because sample sizes were too low for accurate quantification of the incidence of transfusion*

^{10}

^{–}

^{16}). Our approach relies on the median EBL for the procedure only to the extent that this identifies procedures with minimal EBL. Then, the observed incidences of transfusion are used for the MSBOS recommendation.

^{30}Recently, Katz

*et al.*evaluated preoperative testing, recommended that hospitals screen for unnecessary tests, and recommended further investigation whether the benchmarked value to compare providers can be the percentage of patients with at least one unnecessary test.

^{1}Tsai's and Polk's subsequent Letter to the Editor emphasized the need to reduce variability in preoperative testing among providers.

^{31}However, the cost of implementing lean six sigma management processes to reduce unnecessary tests is mostly, if not entirely, a fixed cost (

*i.e.*, same whether one targets one or many preoperative tests). Thus, the economic value of implementing processes to reduce the incidence of unnecessary preoperative tests is increased by scientific understanding of when use of each test is appropriate.

^{31}

*e.g.*, 10 ml) that could not possibly have been measured (at least in adults), or skip the entry. The MSBOS should be used as a recommendation, with treatment of each patient based on clinical information (

*e.g.*, the patient's preoperative hemoglobin and/or recent administration of anticoagulants).

*versus*having a patient undergoing a procedure with a very low incidence of transfusion who has large blood loss and O-negative blood is used.

*e.g.*, major thoracic surgery), erythrocyte transfusion of patients with 1 or 2 units provides no clinical benefit or causes harm.

^{32}

^{–}

^{37}These findings may substantially reduce the percentage of patients receiving transfusion, because, in most circumstances, only 1 or 2 units are administered. A consequence of such a change would be that the transfusion threshold for the median EBL of 50 ml would progressively be larger and more cases and procedures would no longer be included in the MSBOS for type and screen. We think that the benefit of our study is the creation of a framework for taking advantage of such gains in the future. Fortuitously, our framework (table 2) turns out to be simple statistically and should be easy for other hospitals to apply.

*P*(transfusion 1 unit or more) = 1 −

*P*(transfusion with 0 units) = 1 − exp (−1 × mean) = cumulative chi-square distribution with 2 degrees of freedom and parameter 2 × mean.

^{25}Cited Here...

*i.e*., 1 unit or more). Cited Here...

*i.e.*, managerial decision making), rare types of procedures are of much larger importance.

^{10},

^{13}

^{–}

^{16}First, for most decisions what matters is whether an operating room there contains at least one case of a rare type of procedure, which can be much larger probability than the incidence of any one case being of a rare type of procedure.

^{15},

^{16}Second, for case duration prediction, what matters is whether a combination of procedure, surgeon, and type of anesthetic is rare, not just the procedure.

^{26}This markedly increases in the incidence of cases that are of rare combinations. Cited Here...

*e.g*., 20 ml). Our use of 0 ml for missing values matches this decision. Consistent with this model for behavior, among procedures with median EBL less than 50 ml

*versus*50 ml or more from the anesthesia data, the operating room information system was missing the EBL for 88%

*versus*21% of cases, respectively. Classification of the 1,253 procedures as having median EBL less than 50 ml

*versus*50 ml or more was identical to that obtained using the anesthesia data. This was despite the operating room entry differing from the anesthesia entry for 54% of the cases with an EBL entered by the anesthesia provider. Cited Here...

^{10}

^{–}

^{13}We classified each case by its primary procedure code, since only 3.7% of records had another procedure and just 1.3% of such procedures had 19 or more cases (n = 32/2,443). Still, we could study these n = 32 combinations since each had n sufficiently large for median EBL to be insensitive to outliers. Cited Here...

*per se*, which equals 99.2%. This value of 99.2% seems high. However, in the appendix (50 ml or more median EBL less than 500 ml) where only 37 procedures are studied, the lower 95% confidence limit for the concordance equals 92.2%, which seems low. Cohen's kappa would take into account the random probability of agreement. However, neither confidence limit is Bayesian, taking into account our understanding of the reason for the concordance. Our local conclusion was that for median EBL less than 50 ml, the sample size was sufficiently large for clinical implementation. In contrast, for median EBL of at least 50 ml, we consider the value of the work to be its scientific consistency with that for median EBL less than 50 ml. The latter results are shown in the Appendix only. Cited Here...

*e.g*., training and testing dataset) because our results cannot be affected since we use the median EBL. That would not hold, though, for small n. Cited Here...

^{−2}, X

^{−1}, X

^{−0.5}, ln(X), X

^{0.5}, X

^{1}, and X

^{2}. Averaging among hemoglobin categories, the deviance was least for a linear function of the median EBL. Generally, hemoglobin would not be checked for patients when expected to be normal. Among patients with preoperative hemoglobin 12.0 g/dl or larger, the deviance was least for a linear function of the median EBL for all but the hemoglobin category of 13.0 to 13.9 g/dl for which quadratic was significantly less

*P*= 0.0081, but not with correction for the 18 comparisons. Based on these results, we used a linear function of median EBL. Cited Here...