Initial von Willebrand factor antigen values in adolescent females predict future values

Abstract

Introduction:

Diagnosing von Willebrand Disease (VWD) in adolescent females is challenging as menstruation and physiologic stress elevate von Willebrand factor (VWF) laboratory values.

Aim:

To develop a VWF prediction model for adolescent females based on initial VWF results.

Methods:

We identified female patients aged 9 to 21 years with any VWF laboratory test over a 5-year period (2017 thru 2021) at any Texas Children’s Hospital facility. Patient demographics, VWF testing, hemoglobin concentration, serum ferritin, and site of clinical testing were collected (initial and subsequent laboratory evaluations). A Bayesian linear regression model was developed. Prediction intervals were analyzed to identify thresholds for patients in whom repeat testing was unlikely to identify low VWF levels (<50%), consistent with VWD.

Results:

A total of 6,125 adolescent females underwent VWF testing; 1,204 (19.7%) had repeat testing. Based on the prediction model, initial VWF antigen values of 80%, 90%, and ≥100% carried a 92.6%, 96.6%, and ≥98.0% probability of having repeat normal repeat VWF values, respectively. Subjects assessed in outpatient adolescent medicine or gynecology clinics were more likely to have low VWF values compared to those assessed in the acute care setting (p<0.001). Median presenting hemoglobin and serum ferritin were 12.4 g/dL and 13 ng/mL, respectively and were similar in those with normal versus low VWF antigen values.

Conclusion:

Repeat testing in adolescent females whose initial VWF antigen values are ≥90% is unlikely to identify additional patients with VWD. Iron deficiency screening should be performed in all adolescent females.

Introduction

Von Willebrand Disease (VWD) is the most common inherited bleeding disorder, affecting up to 1% of the general population.[1, 2] In many adolescents and young adults, heavy menstrual bleeding (HMB) is the initial manifestation of an underlying bleeding disorder, and most common reason for hemostatic laboratory testing.[3, 4] Given the prevalence of VWD, the American College of Obstetricians and Gynecologists recommends that VWD should be considered in all patients with HMB and evaluated with laboratory testing.[5] Interpretation of von Willebrand Factor (VWF) laboratory values in adolescent female patients, however, is challenging.

Physiologic stress, inflammation, exogenous estrogen, iron deficiency, and acute illness elevate VWF levels.[6–8] Likewise, acute HMB causes elevated VWF laboratory values and results in an inaccurate assessment of the patient’s baseline VWF values. Therefore, in any patient being evaluated for VWD, repeat testing is often required to rule out or confirm the diagnosis of VWD.[9] Data to guide the interpretation of VWF values obtained specifically in the adolescent female population are limited. The aims of our study were to understand the VWD diagnostic testing patterns in adolescent females within the Texas Children’s Hospital (TCH) network over a five-year span and to develop a prediction model to determine whether repeat testing was required to rule-out the diagnosis of VWD, based on initial VWF testing results in this population.

Materials and Methods

A query of the TCH electronic medical record system identified all female patients ages 9 to 21 years who had any VWF testing performed at a TCH facility (across the greater Houston Metroplex) over a five-year period (January 1, 2017 to December 31, 2021). Laboratory data collected from the initial and subsequent laboratory testing visits included VWF antigen, VWF activity (ristocetin cofactor [RCo] or glycoprotein Ib [GPIb] assay), factor VIII (FVIII) activity, hemoglobin concentration, and serum ferritin. All VWF diagnostic testing ordered by a TCH facility was gathered, including VWF tests performed at a TCH laboratory and those performed at an external laboratory. Patient demographics and the site of laboratory collection were gathered and categorized as either an acute care facility (emergency center or inpatient facility), outpatient adolescent medicine or gynecology clinic, outpatient hematology clinic, or other outpatient facility. Due to the change in VWF activity assays performed across TCH facilities over the study time frame, VWF antigen values were used to compare individual patient values across the cohort. The indication for VWF testing, family, and clinical histories (including bleeding symptoms) were not included in data collection. Additional non-specific laboratory screening tests, including activated partial thromboplastin times and prothrombin times, were not collected.

Females who had VWF laboratory testing performed between 2010 and 2016 at a TCH facility were excluded from the cohort to ensure that repeat VWF testing was not mistaken as an initial VWF test. Patients who had multiple VWF tests performed within a 7-day period were manually reviewed. VWF laboratory values obtained following a dose of desmopressin or plasma derived or recombinant VWF products were excluded, including those who underwent an outpatient pharmacokinetics study or treatment related to dental or surgical procedures. Repeat VWF testing was performed at the discretion of the treating clinician.

The joint American Society of Hematology (ASH), International Society on Thrombosis and Haemostasis (ISTH), National Hemophilia Foundation (NHF), and World Federation of Hemophilia (WFH) guidelines on the diagnosis of VWD state that a VWF level of <50% in the setting of abnormal bleeding is consistent with type 1 VWD.[10] Therefore, we defined VWF antigen values <50% as abnormal, with the presumption that adolescent females undergoing laboratory evaluation for VWD were doing so in the setting of HMB. This study was approved by the Baylor College of Medicine Institutional Review Board.

Model development

We developed our prediction model according to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) framework[11] (see Appendix A for the TRIPOD checklist). The statistical model allowed us to account for repeat testing and produce probabilistic predictions for the outcomes across a range of clinical contexts and risk factors. The outcome variable of interest was predicting future VWF antigen levels based on previous VWF antigen levels and other predictors. Only subjects with two or more VWF levels were retained in the model. Candidate predictor variables included VWF antigen levels, age, sex, body mass index (BMI), hemoglobin concentration, serum ferritin, FVIII activity levels, the location of lab draw classified as either acute care (VWF labs drawn from the emergency center or inpatient departments) or non-acute care (outpatient clinics), and the total number of VWF antigen levels assessed. We planned to fit all continuous variables with restricted cubic splines using 3–5 knots for each. We calculated a required sample size of at least 716 observations to predict VWF as a continuous outcome with a theoretical R² of 0.2 and 20 total parameters. Bayesian methods were used for all modeling steps (Appendix B).

Because the variables for hemoglobin concentration, serum ferritin, FVIII activity, and BMI were missing for >50% of the observations, multiple imputation with model averaging was used to fit a model that included all predictors. We also fit a model with the predictors with no missing information that included VWF antigen, age, location of lab draw, and total number of VWF labs obtained. The models were compared using Pareto smoothed importance sampling leave-one-out (LOO) cross-validation (CV). LOO-CV characteristics and other performance metrics were similar between the two models (Appendix B - Table S1). We chose the simpler model using only age, sex, lab draw location, lab number, and initial VWF antigen level as the final model because there was no degradation in CV performance, and the simpler model could be more readily implemented when hemoglobin concentration, serum ferritin, FVIII activity or BMI were unavailable. See Appendix C for VWF predictions by age, number of previous VWF lab draws, and previous evaluation setting.

Three candidate models were then compared: multivariable linear regression, multivariable linear regression with a logarithmic transform for the outcome and predictor VWF antigen levels, and semiparametric ordinal regression. The multivariable linear regression model with the logarithmic transformation demonstrated superior performance characteristics and was chosen as the desired model form. This model was used to produce predictions of VWF antigens and predictive performance was evaluated. All statistical analysis was performed in R version 4.2.2.[12] The “rstan”, “brms”, “rmsb”, “hmisc”, “bayesplot”, and “loo” packages were used for model development, evaluation, and analysis.[13–17]

Results

A total of 6,125 adolescent females (median age 14 [IQR 12, 16]) had VWF diagnostic testing at a TCH facility during the study period (Table 1). The median presenting VWF antigen, hemoglobin concentration, and serum ferritin were 89%, 12.4 g/dL, and 13 ng/mL, respectively. Repeat VWF testing was performed in 1,204 (19.7%) patients, at the discretion of the treating clinician. Females that underwent repeat VWF testing had lower initial VWF antigen, activity, hemoglobin concentration, and serum ferritin values compared to those who only had one VWF laboratory test (Table 1). Von Willebrand antigen testing was performed ≥3 times in 314 (5.1%), ≥4 times in 98 (1.6%), and ≥5 times in 29 (0.5%). Of the 1,204 patients who had repeat VWF testing, 1,037 (86%) had normal initial and repeat values, 75 (6.2%) had low initial and repeat values. The remaining patients had discrepancies between initial and repeat testing. Forty-seven (3.9%) had low initial values that were normal on repeat testing while 45 (3.7%) had normal initial values that were low on repeat testing (Table 2). Initial median VWF antigen values of these subgroups were 88%, 42%, 45%, and 54%, respectively.

Table 1.

Demographic and Clinical Characteristics of Adolescent Female Patients Evaluated for von Willebrand Disease Based on Number of VWF Testing Performed

Characteristic	Multiple VWF tests, N=1,204		Single VWF test, N=4,921		p-value1
Age in years, median (IQR)	14	(12, 16)	14	(12, 16)	<0.001
Race/Ethnicity, n (%)
American Indian	4	(0.4%)	12	(0.2%)
Asian & Pacific Islander	52	(4.3%)	221	(4.5%)
Black (Non-Latinx/Hispanic)	206	(17.1%)	853	(17.3%)
Latinx/Hispanic	566	(47.0%)	2,040	(41.4%)
White (Non-Latinx/Hispanic)	347	(28.8%)	1,553	(31.7%)
Unknown	29	(2.4%)	242	(4.9%)
Body mass index, median (IQR)	23	(20, 27)	23	(20, 28)	0.8
Location of evaluation, n (%)					<0.001
Acute care*	260	(22%)	569	(11%)
Adolescent Medicine or Pediatric Gynecology	435	((36%)	2,298	(47%)
Hematology	208	(17%)	594	(12%)
Other outpatient clinic	301	(25%)	1,460	(30%)
Laboratory values measured at presentation, median (n)^ (IQR)
VWF Antigen (%)	78 (1,204)	(59, 121)	94 (4,921)	(74, 120)	<0.001
VWF Activity (%)	75 (1,161)	(55, 111)	89 (4,728)	(71, 116)	<0.001
Factor VIII (%)	139 (517)	(90, 226)	133 (1,441)	(102, 180)	0.4
Hemoglobin concentration (g/dL)	12.1 (510)	(10.4, 13.1)	12.5 (1,855)	(11.6, 13.3)	<0.001
Serum ferritin (ng/mL)	12 (646)	(5, 23)	14 (2,747)	(7, 24)	<0.001

VWF = von Willebrand Factor

¹Wilcoxon rank sum test; Pearson’s Chi-squared test; Fisher’s exact test

^*Acute care = emergency center or inpatient setting

^{^}(n) represents the subtotal for whom laboratory values were available

Table 2.

Clinical Characteristics of Adolescent Females by Initial and Repeat VWF Testing

Total Cohort (N=1,204) Clinical Characteristic	Normal initial, Normal repeat VWF, N=1,037 (86.1%)		Low initial, Low repeat VWF, N=75 (6.3%)		Low initial, Normal repeat VWF, N=47 (3.9%)		Normal initial, Low repeat VWF N=45 (3.7%)		p-value1
Age in years, median (IQR)	14	(12, 15)	15	(13.5, 17)	15	(13, 16)	14	(13, 16)	<0.001
Race/Ethnicity, n (%)
American Indian	4	(0.4%)	0	(0%)	0	(0%)	0	(0%)
Asian & Pacific Islander	45	(4.4%)	3	(4.0%)	2	(4.2%)	2	(4.4%)
Black (Non-Latinx/Hispanic)	186	(18%)	8	(10.7%)	3	(6.4%)	9	(20%)
Latinx/Hispanic	479	(46%)	35	(46.7%)	31	(66%)	21	(46.7%)
White (Non-Latinx/Hispanic)	301	(29%)	26	(34.6%)	8	(17%)	12	(26.7%)
Unknown	22	(2.2%)	3	(4.0%)	3	(6.4%)	1	(2.2%)
Body mass index, median (IQR)	23	(20, 27)	22	(20, 26)	23	(20, 26)	24	(20, 27)	>0.9
Location of evaluation, n (%)									<0.001
Acute care*	248	(24%)	3	(4%)	3	(7%)	6	(13%)
Adolescent Medicine or Gynecology	345	(33%)	36	(48%)	27	(57%)	27	(60%)
Hematology	178	(17%)	17	(23%)	8	(17%)	5	(11%)
Other outpatient	266	(26%)	19	(25%)	9	(19%)	7	(16%)
Baseline Laboratory Values, median (n)^ (IQR)
VWF Antigen (%)	88 (1,037)	(65, 131)	42 (75)	(37, 46)	45 (47)	(43, 48)	54 (45)	(51, 62)	<0.001
VWF Activity (%)	82 (1,000)	(61, 117)	42 (73)	(33, 47)	44 (46)	(38, 49)	55 (42)	(47, 61)	<0.001
Factor VIII (%)	159 (455)	(101, 238)	58 (32)	(51, 70)	75 (17)	(66, 81)	102 (13)	(79, 116)	<0.001
Hemoglobin concentration (g/dL)	12.1 (435)	(10.2, 13.1)	12.8(33)	(11.2, 13.3)	12.3(23)	(11.8, 12.8)	12.4(19)	(10.9, 13.5)	0.2
Serum ferritin (ng/mL)	12 (545)	(5, 23)	14 (42)	(9, 24)	11 (29)	(6, 15)	10 (30)	(5, 20)	0.4

VWF = von Willebrand Factor

^*Acute care = emergency center or inpatient setting

¹Kruskal-Wallis rank sum test; Pearson’s Chi-squared test;

^{^}(n) represents the subtotal for whom laboratory values were available

Females with a normal initial VWF antigen that underwent repeat VWF testing were younger compared to those with abnormal initial values with median ages of 14 and 15 years, respectively (p<0.001). There was no difference in presenting hemoglobin concentration or serum ferritin values between the four subgroups. All patient subgroups with repeat VWF testing had a median initial ferritin value of less than 15 ng/mL, consistent with iron deficiency. Patients with normal initial VWF testing were more likely to have been tested in an acute care setting. Those with low initial VWF antigen values were more likely have been tested in an outpatient adolescent medicine or gynecology clinic (Table 2, p<0.001).

While it is the authors’ practice to obtain FVIII activity values when performing laboratory evaluation for VWF, a low proportion of laboratory evaluations within the cohort, 42% of multiple VWF test group and 29% in the single VWF test group, contained both VWF antigen and FVIII activity. This likely reflects different ordering practices of providers across the community and different FVIII reporting in sendout tests (e.g. reported as a “miscellaneous” or “other” test result, which was not included in the laboratory data collection).

Prediction Model

The Bayesian multivariable model demonstrated good predictive performance. The Bayesian R², a measure that estimates the proportion of the variance of the outcome explained by the model, for the full dataset was 0.70 (95% Compatibility Interval [CI]: 0.69–0.71). The LOO-CV R², an estimate of performance on future observations, was 0.69 (95%CI: 0.65–0.73). The posterior predictive distribution reproduced the distribution of observed outcomes appropriately (Figure 1A). The distribution of skew and kurtosis statistics of the posterior distribution closely approximated the skew and kurtosis of the observed data (Figures 1B and 1C). These aspects of the posterior predictions indicated appropriate distributional properties for our prediction task. Further performance data are available in Appendix B.

Figure 1.

Posterior predictive and observed distributional characteristics. For all panels, the light blue graphs indicate the value of one draw from the posterior predictive distribution. Dark blue lines indicated the statistics from the observed outcomes. A: Density graph of predicted von Willebrand Factor antigen levels. B: Distribution of skewness. C. Distribution of Kurtosis.

The marginal relationships between each predictor and VWF antigen level are shown in Figure 2. The previous VWF antigen level (Figure 2A), age (Figure 2B), and number of evaluations from the initial VWF antigen level (Figure 2C) were positively associated with the predicted VWF antigen level. Previous evaluation in the acute care setting was associated with a decreased predictive VWF antigen level compared to previous evaluations outside the acute care setting (Figure 2D). To produce these marginal relationships for each variable in the multivariable model, the other covariate values had to be specified. We set the values for previous VWF antigen and age to the mean values in the dataset. We set the number of labs from the initial evaluation to the first value and the previous department to acute care. These settings for the categorical variables produced the lowest predicted values for the VWF antigen level, which provided the highest predictions that the VWF antigen levels would fall to less than 50%.

Figure 2.

Predicted von Willebrand factor (VWF) levels by variable value from the multivariable prediction model by previous VWF antigen level (A), age in years (B), number of lab draws after the first evaluation (C), and the location of the previous evaluation (D). For each of the predictions, the continuous non-displayed covariates were set to mean values (previous VWF antigen level = 96%, age = 14.4 years), and the discrete covariates were set to reference levels (number of VWF antigen levels from first evaluation = 1, location = acute care). The blue line shows the median value, and the gray areas or bars represent the 95% compatibility intervals.

The predicted VWF antigen levels compared to the observed values in the dataset by previous VWF antigen level are shown in Figure 3A. The cumulative probability that the predicted VWF antigen value would be higher than 50% for each value of the previous VWF level is also displayed (Figured 3B). The covariate settings for the other covariates are the same as in the marginal relationships. Patients with an initial VWF antigen of 50% had a 47.5% probability of having repeat VWF antigen of 50% or higher. Initial VWF antigen vales of 80%, 90%, and ≥100% had a probability of normal repeat testing of 92.6%, 96.6%, and ≥98.0%, respectively (Table 3).

Figure 3.

A: Predicted von Willebrand factor (VWF) antigen levels by previous VWF level. Blue bars indicate prediction intervals. Gray points represent median prediction value. B: The probability that the predicted VWF level will be greater than 50% by previous VWF antigen level. For all predictions from the multivariable model, age was set to the mean value (14.4 years), the number of values from first value was set to 1, and the previous department was set to “acute care”.

Table 3.

Probability of Normal VWF Repeat Testing Based on Initial VWF Value

Previous VWF Antigen	Probability of normal (≥50%) VWF value on repeat testing
≤20%	0.0%
30%	3.6%
40%	20.5%
50%	47.5%
60%	70.8%
70%	85.1%
80%	92.6%
90%	96.6%
100%	98.0%
110%	99.1%
≥120%	≥99.4%

VWF = von Willebrand Factor

Predicted probabilities produced from the multivariable model with the covariates set at the average age (14.4 years), 1^st repeat lab after initial evaluation, and previous evaluation in the acute care setting.

Discussion

Over six thousand adolescent females were evaluated for VWD at a TCH facility over a five-year period; approximately twenty percent underwent repeat testing. Our data suggest that adolescent females who undergo VWD diagnostic testing do not require repeat testing if their initial VWF antigen value is 90% or higher. We additionally found that median ferritin values for the overall cohort, and subgroups of females with differing VWF values, were consistent with iron deficiency.[18, 19]

A retrospective study aiming to characterize the ability to detect VWD in adolescents with acute HMB demonstrated a reduction in VWF antigen values of approximately 25% in an outpatient follow-up compared to their emergency department presentation for the 39 patients that underwent repeat testing. In this study, an initial VWF antigen >100% carried a 93.2% negative predictive value (NPV) for VWD.[20] A separate cohort of male and female pediatric and adolescent patients (ages 0 to 18 years) by Doshi et al. found that VWF antigen cutoffs of 75% and 100% carry NPVs for VWD of 94.1% and 96.6%, respectively.[21] Whether findings from that study can be applied to adolescent females is unclear, as only 18.2% of patients in the cohort underwent VWD diagnostic testing due to HMB.[21]

Many adolescent females have initial hemostatic testing performed upon presentation to an acute care setting for bleeding symptoms such as HMB. Given the physiologic stress that may be present in patients seeking acute care, VWF diagnostic testing obtained in such settings has previously been of limited diagnostic utility, as the results are unlikely to represent an individual’s baseline VWF levels. The results of our prediction model, however, demonstrates that initial VWF antigen values of 90% or higher can be used to rule out VWD (with a 3.4% false negative rate), and individuals with VWF antigen values of 50–90% should undergo repeat VWF testing. The VWF antigen threshold of 90% balances the risks of a false negative with the discomfort and increased resource requirements associated with repeat laboratory assessment. Practitioners should use the probabilities listed in Table 3 to decide the most reasonable threshold for their clinical and patient population. As an example, a higher VWF threshold for re-testing and lower false negative rate would be required for an adolescent female with a positive bleeding assessment tool score, severe anemia, and multiple family members with bleeding symptoms compared to a female with isolated heavy bleeding shortly after menarche and negative family history.

VWD is common in adolescent females with HMB. One prospective study in adolescents with HMB who were referred to an outpatient bleeding disorder clinic observed a VWD incidence of 36%, according to the 2021 ASH ISTH NHF WFH guidelines on the diagnosis of VWD.[10, 22] We identified a higher prevalence of abnormal VWF values when testing occurred in an outpatient adolescent medicine or gynecology clinic. Potentially, families affected by an inherited bleeding disorder (whether diagnosed or undiagnosed) normalize HMB and seek outpatient care for evaluation. In contrast, families not affected by VWD may be more likely to view HMB as an emergent symptom and present for emergent medical care. In any clinical setting, it is critical to consider the diagnosis of VWD for patients presenting with HMB, as VWD has historically been underrecognized in this age group.[1, 4, 23, 24]

The widespread presence of iron deficiency within our cohort stresses the importance of screening this population by obtaining a serum ferritin and treating with iron therapy when iron deficiency (ferritin <20 ng/mL) is present.[18, 19] Similar to under recognition of VWD, the association of HMB with iron deficiency is often overlooked yet has important clinical implications given its associations with poor concentration, fatigue, restless legs, and decreased school performance in adolescent females.[25–29] Earlier detection and treatment would also prevent progression to anemia and related complications. A previous study of adolescent females with HMB admitted for severe anemia revealed that only 39% had any iron laboratory assessment and 37% of admissions did not initiate iron therapy.[30] Screening with a hemoglobin or complete blood count alone is inappropriate as less than half of adolescent females may be anemic or microcytic.[31] Our study was consistent with these findings as the median hemoglobin for the entire cohort was 12.4 g/dL (i.e. within normal range) while the median ferritin was low at 13 ng/mL.

The limitations of this study include the inherent biases in its retrospective nature. We analyzed repeat VWF values available in the dataset. This may represent a form of referral bias as patients who had repeat testing may represent patients with at higher risk for VWD. Prospective validation of these predictions is needed. Initial and repeat diagnostic testing was performed inconsistently in the cohort, reflecting the clinical practices of practitioners across a variety of different specialties and settings. We chose to compare VWF antigen values rather than VWF activity across the cohort. For those patients who had a more pronounced reduction of VWF activity than antigen, consistent with Type 2 VWD, using the antigen value alone may have inadvertently classified them as “normal.” However, the VWF antigen and activity values in our cohort were highly correlated, making this less likely. Not all VWF diagnostic testing in the cohort was performed at an onsite coagulation laboratory. We did not include the site of laboratory processing as a variable in our prediction model, as frequently hematologists are required to interpret these off-site VWF values. Our prediction model is therefore applicable in a real-world setting with a heterogeneity of coagulation laboratories used. The predictive model’s high performance also makes it unlikely that adding the location of laboratory processing as an additional predictor variable would impact the predicted repeat VWF value. Finally, we did not include the indications for VWF laboratory testing or patient bleeding histories in our data collection or predictive model development. As is the case, the cohort likely includes individuals who presented for reasons other than HMB (e.g., epistaxis or surgical bleeding).

This study has many strengths. Our sample size was sufficient to flexibly model the outcome with the required degree of precision. The prediction model performed well and yielded a high Bayesian R². Our use of statistical modeling represents a methodological advance from previous studies.[32] Through modeling, we demonstrated how VWF levels vary across a range of clinical scenarios and risk factors to show how the performance of a threshold may change in a given context. The probabilistic predictions allow providers to determine the most appropriate threshold for repeat testing based on the clinical context, costs of the decision, and other information available to them. It also allows providers to interpret VWF values obtained during times of stress, including in acute care settings, without the need to repeat testing in all patients.

Conclusion

We found that adolescent females are unlikely to have a diagnosis of VWD, or require repeat VWF testing, if their initial VWF antigen is 90% or higher. All adolescent females undergoing evaluation for a bleeding disorder should be evaluated for iron deficiency with serum ferritin screening. Future studies are needed to validate the VWF repeat testing cutoff in this patient population and others and to evaluate the effect of iron status and severe anemia on repeat VWF levels in adolescent females.