Desmopressin Dosing in Children Using Real-World Data and Pharmacokinetic/Pharmacodynamic Model Simulations

Abstract

Background

Variability in pediatric dosing of desmopressin (ddAVP) in AVP-deficiency (AVP-D) is well-documented but dosing recommendations are limited. This study evaluates and optimizes ddAVP dosing regimens in children with AVP-D using pharmacokinetic and pharmacodynamic (PK/PD) simulations.

Methods

Retrospective electronic health record review was done to identify children (< 18 years) with AVP-D on ddAVP evaluated in the outpatient setting using ICD 9 and 10 codes. A previously developed PK/PD model from Michelet et al was used to simulate ddAVP concentrations and urine rates based on a child’s age and ddAVP dose. The effects of demographic characteristics (age, weight, etc.) on dose and urine rate were investigated through simulations to optimize doses of ddAVP for children who were wet overnight.

Result

A total of 276 dosing records were identified among 53 children with AVP-D. Simulations indicated that in children under 5 years of age who were wet overnight, increasing the outpatient dose to 50 mcg was predicted to decrease urine rate to a pattern similar to those who remained dry.

Conclusion

An initial outpatient dose of at least 50 mcg for children between 1 to 5 years of age would improve efficacy of ddAVP.

Introduction

AVP-deficiency (AVP-D), formerly known as central diabetes insipidus is a rare endocrine condition in which the lack of endogenous arginine vasopressin (AVP) leads to excessive urine production and output ¹. Its underlying etiology varies but frequently involves either loss or destruction of the hypothalamo-neurohypophysial unit. Management of AVP-D is known to be difficult. This is especially true in children as it can be difficult to deliver small and precise doses from a given formulation ^2-5. Desmopressin (ddAVP) is commonly used to treat AVP-D and is available in different formulations including subcutaneous, intranasal, oral lyophilizate, and oral tablets, with oral tablets used most commonly in children⁶. While ddAVP has proven to be efficacious in preventing polyuria, it has a narrow therapeutic index due to the mechanism and duration of action. This places individuals treated with ddAVP at risk of hyponatremia from excessive water retention or hypernatremia from excessive water clearance, both of which can be life-threatening ². One of the most pressing challenges in managing AVP-D in children is the lack of specific dosing guidelines. The current labeling for ddAVP is extremely limited. Dosage recommendations exist only for children 4 years of age or older, and suggest a starting dose of 50 mcg with non-specific instructions to “titrate to effect” ⁷. However, we have observed a wide range of prescribing practices and have found that clinicians often start at a lower dose. The lack of dosing guidelines puts children, especially those under 4 years, at risk of under and overdosing and prolongs the duration over which children receive the incorrect dose. While the heterogeneity and variability in pediatric doses of ddAVP is well-documented, there have not been any advances in dosing recommendations since its initial approval^5,6,8.

Pharmacokinetic/Pharmacodynamic (PK/PD) modeling is a validated method of estimating the relationship between drug dose and effect through mathematical models. These models are also able to identify variables such as demographic characteristics which are responsible for differences in a sub-population’s drug concentrations and or response^9-11. This method has been used to optimize dosing of drugs with narrow therapeutic windows in oncology as well as infectious diseases ^12-14. Recently, Michelet et al developed a PK/PD model for oral ddAVP tablets in children with nocturnal enuresis that describes the duration of staying dry or anuria after a ddAVP dose ¹⁵. This study aims to utilize the existing PK/PD model for ddAVP from Michelet et al to analyze dosing of ddAVP among children with AVP-D to elucidate patterns in dose effectiveness and develop dosing recommendations.

Methods

Clinical Data

This was a single center study conducted at the University of California, San Francisco which included children seen at both San Francisco and Oakland campuses. This study was reviewed and deemed exempt by University of California, San Francisco’s Institutional Review Board. Clinical data were extracted retrospectively from the electronic health record (EHR) system using ICD codes for AVP-D (ICD10: E23.2, ICD 9: 253.5) from years 2012 to 2022. Inclusion criteria were age less than 18 years and the consistent use of daily oral ddAVP tablets without discontinuation. Participants with partial or “resolved” DI were excluded, as well as participants with known chronic renal impairment. Dosing records, participant demographics, baseline characteristics, as well as anthropometrics (height, weight, body mass index (BMI) were obtained via chart review. Only outpatient dosing records were selected to ensure ddAVP doses reflected daily use under steady state.

Statistical Analysis/Model building

Statistical analyses including t-test, multivariate logistical regression, and summary statistics were performed with STATA SE 17. PK/PD models as well as model simulation outputs were generated using the PKPDsim package in R. The PK/PD model was adapted from Michelet et al, which is a one compartment model with linear oral absorption and elimination. A sigmoidal Max model described the relationship between ddAVP concentration and urine rate. Age is a significant covariate on bioavailability and urine production rate¹⁵. The primary PD outcome in the Michelet et al model is urine output in ml/hr whereby the duration of action of ddAVP is reflected by the time an individual is anuria/oliguria. As time in anuria/oliguria is difficult to interpret from free text in EHRs, the ability to remain dry without urine output overnight (treated as a binary measure yes/no) was used as a surrogate marker for efficacy instead. The continuous urine rate was translated to a binary outcome by chart review on dosing adequacy. This is possible as urine output in Michelet et al’s model have rapid changes from baseline to near zero once ddAVP is given and rapid return of urine production once ddAVP is “off”. This description of rapid changes in urine output is analogous to the clinically observed effects of ddAVP whereby urine output is quickly shutdown after a dose is given and a short window for urine “breakthrough” exists in between ddAVP doses. In this analysis, the dose of interest was the nighttime dose, which was assumed to be taken in the fed state. Each of the 53 individuals included in the dataset were simulated over 100 repetitions.

Results

A total of 276 dosing records for oral ddAVP tablets were identified among 53 individuals from EHRs. The demographic characteristics are described in Table 1 but in brief, 25 (47%) of the participants were female, and the average age was 11.8 years with range of 1 to 17.8 years. The average number of dosing records per individual was 4.28 and ranged from 1 to 18 records per individual due to different lengths of follow up. 41.5% of our cohort self-identified as White. The causes for AVP-D are listed in Table 2 of which the most common are central nervous system tumors such as craniopharyngioma (22.6%), Langerhans cell histiocytosis (18.9%), and germinomas (17%). This distribution of underlying etiology is similar to previously reported studies ^16-18.

Table 1.

Baseline characteristics of pediatric patients with cDI at time of data extraction (53 individuals, 276 dosing records)

Age in years, median (IQR)	8.6	(5.5-11.8)
Sex, n (%F)	24	(47%)
Weight in kg, median (IQR)	32	(20-54.5)
Height in cm, median (IQR)	126	(109-147)
BMIz, mean (SD)	1.7	(2.1)
Panhypopituitarism, n (%)	47	(88%)
Race/ethnicity, n (%)
White	22	(41.5%)
Latinx	16	(30.2%)
Asian/Pacific Islander	9	(17.0%)
Black	3	(5.7%)
Other	3	(5.7%)

Table 2.

Underlying diagnosis for cDI in retrospective pediatric patient cohort (53 individuals)

Underlying diagnosis	n	(%)
Craniopharyngioma	12	(22.6)
Langerhan cell histiocytosis	10	(18.9)
Germinoma	9	(17)
Septo-optic dysplasia	7	(13.2)
Congenital DI	4	(7.5)
Structural defect	4	(7.5)
Trauma	3	(5.7)
Genetic	1	(1.9)
Isolated	1	(1.9)
Meningitis	1	(1.9)
Unclear	1	(1.9)

Supplementary Figure S1 displays the different ddAVP doses based on age. There is high variability in doses spread across different age groups. Supplementary Figure S2 shows changes in ddAVP dose for each individual participant over time. While there is a general increase in dose with age, many participants did not have any changes in ddAVP dose for years.

There is a positive correlation between ddAVP dose and weight (R² = 0.1774), height (R² = 0.2386), and age (R² = 0.1876), with small R² values suggesting a weak correlation (Supplementary Fig. S3). When stratified by wet or dry status overnight as a surrogate marker for drug efficacy, those who remained dry were on a higher dose of ddAVP (p = 0.008), older (p = 0.016), and had a larger model simulated area under the concentration time curve (AUC; 0.0038) by t-test. Height and weight were not significantly different between dry and wet groups (Supplementary Tab. S1). Multivariate logistic regression showed weight (OR = 0.97, p = 0.019), dose (OR = 1.01, p = 0.007), and age (OR = 1.33, p = 0.01) as significant predictors for dry overnight status (Supplementary Tab. S2).

When stratified by age, the model-based simulations revealed much lower exposures in children less than 5 years old (Supplementary Fig. S4, S5; 59 dosing records among 15 individuals). Most children under 5 years old who are wet overnight received ddAVP doses less than 50 mcg which is significantly lower than those who are dry (Fig. 1; p = 0.0001). Higher doses ranging from 60-90 mcg were simulated in wet children under 5 to investigate which dose provided similar ddAVP exposures and urine rates (Fig. 2, 3). The simulations revealed that in order to match the typical urine rate of dry children a dose between 70 and 80mcg is needed.

Fig 1.

Distribution of ddAVP doses in children under 5 years of age. The dashed lines represent the mean dose given for wet and dry children respectively.

Fig 2.

Simulated diuresis rates over time in children who were wet overnight under 5 years of age. The original profile of those who were wet and dry overnight are included for reference. In the wet overnight population doses of 30-50 mcg were simulated. The dashed line is the median of the population and the shaded regions are the 95% confidence intervals.

Fig 3.

Simulated diuresis rates over time in children under 5 years of age who were wet overnight. The original profile of those who were dry overnight is included for reference. In the wet overnight population, doses of 60-90 mcg were simulated. The dashed line is the median of the population and the shaded regions are the 95% confidence intervals.

Discussion and Conclusion

Based on our simulations, an outpatient dose of 70 mcg daily oral ddAVP is needed to match urine rate profiles to children who remain dry overnight. However, for some children this would result in more than doubling their initial dose. Given the narrow therapeutic index of ddAVP a more conservative dose of 50 or 60 mcg may be more appropriate in the outpatient setting. As children with AVP-D under the age of 1 are frequently managed with thiazide diuretics, the dosing recommendation of 50mcg of ddAVP is most appropriate for children between 1 to 5 years of age based on our results. To our knowledge, this is the first study to suggest doses for pediatric ddAVP in recent years. Although we were unable to provide more specificity or age-based dosing, the results clearly displayed that younger children are underdosed. ddAVP dosing and AVP-D is often the most difficult to manage in young children and recommendations for an effective starting dose in the less than 5 years of age group are much needed. A safe and effective dose for ddAVP initiation can save time, effort, as well as frustration and uncertainty that can come with titrating ddAVP ^1,2,19.

While the dosing ranges in our population were similar to that of a recently described study^8,20,21 there was a large amount of variability in dose and dose changes over time (Supplementary Fig. S1, S2). Our findings not only capture the well-known heterogeneity in ddAVP dose but interestingly highlight the variability in dose changes where some participants went years without having a dose adjustment (Supplementary Fig. S2). We also showed that ddAVP dose generally increases with age and weight although the correlation is weak (R² values 0.1876 and 0.1774, respectively). This phenomenon can potentially be explained by the fact that AVP-D can be managed with fluids even with inadequate or no ddAVP^22-24. Therefore, it is possible that individuals compensate for ineffective ddAVP dosing by adjusting their fluid intake leading to long periods of time without a dose adjustment. The interplay between ddAVP dosing and fluid intake further underscores the difficulty in determining appropriate ddAVP doses. Another source of dosing variability was found during our review of inpatient admissions, where children frequently had very different responses to similar doses of ddAVP during short hospitalizations with acute illnesses/conditions. This likely highlights variable response to ddAVP while under physiologic stress. Ultimately, inpatient dosing records were not included in our analysis for this reason as variable response to ddAVP in short period of time is not conducive to predictive model building. Lastly review of anthropometric data shows that there is great variability in the range of weight, height, and BMI z scores (−4.5-4, −4.3-3.7, −6-7.1, respectively). These variabilities likely represent the clinical status of this population as they frequently have panhypopituitarism and are undergoing chemotherapy. These factors alone and in concert can have dramatic impacts on ddAVP drug response, making dosing decisions even more difficult.

Children older than 5 years showed a similar trend in higher ddAVP exposures in those who remained dry overnight compared to those who were wet. However, the PD model predicted dry children as having higher urine production rates, the opposite of what we would expect. A possible explanation is that the ability to stay dry overnight relies on both bladder control in addition to suppression of urine production. Therefore, it is possible for older children to stay dry overnight despite higher urine production. This may be a reason that older age is a predictor of being able to stay dry overnight in our multivariate regression model. We did attempt to adjust and normalize urine production output to bladder volume using age and weight-based approaches^25,26; however, this did not appear to change the observed trends. Another possible explanation is that the effect of age on urine production rate included in the model is different in our patient population.

It is notable that in our study, children who were able to remain dry overnight tended to be older and taking higher doses of ddAVP. Interestingly, their weight and height appear to be similar. This further suggests that age is likely the most significant factor in ddAVP dosing as reflected in the Michelet model. Obesity status was distributed equally between the two groups as well. In multivariate regression analysis for the ability to remain dry, we found again that age and dose were positive predictors for ability to remain dry. Interestingly, the same analysis found that those who weigh less are more likely to remain dry overnight. This suggests may suggest that weight, independent of age, is important to ddAVP dosing as well.

We recognize that due to the retrospective nature of this study we were unable to build our own model or to improve or modify previous models, largely because the duration of anuria/oliguria after a dose of ddAVP is not always clearly documented. Additionally, the nature of our EHR system is such that time course of AVP-D diagnosis and ddAVP use difficult to extract and interpret. These are certainly weakness of this study design. It is possible that we would be able to provide further specificity in dosing if data with duration of anuria/oliguria were available. This also speaks to the lack of uniformity in documenting information in clinical encounters. Recent efforts in other endocrine conditions such as type 1 diabetes (T1D) have developed registry programs to establish uniform data collection to inform and answer scholarly questions. It is important to establish similar standards for conditions such as AVP-D, and building a registry or repository of uniform data across institutions will help address research questions.

Many children in our study are receiving ddAVP doses less than those reported in the Michelet et al study¹⁵. One possibility is that the model was developed using data from children with nocturnal enuresis who have intact endogenous AVP secretion. Therefore, higher doses may be required to overwhelm V2 receptors. As children with AVP-D have lack of endogenous AVP, they may need smaller doses to achieve an effect given saturation of V2 receptors.

Overall, our study is novel in that we were able to leverage existing data from EHRs and an existing PK/PD model to gain insight into ddAVP dosing. While this is not a replacement for prospective and dedicated PK/PD studies of ddAVP in children with AVP-D, it highlights the possibility of using available clinical data to improve the dosing specificity of medication and suggests a range of doses to test in future ddAVP clinical trials.

Impact Statement:

• 50mcg is likely a safe initial outpatient dose of oral desmopressin tablet for young children 1-5yrs of age with central Diabetes Insipidus/AVP-Deficiency

• We confirmed that desmopressin doses vary greatly in children with central Diabetes Insipidus/AVP-deficiency

• Real-world clinical data can be leveraged to improve medication dosing in rare diseases

Data Availability Statement

All data generated and analyzed for this study were included in the study. For further information, please contact corresponding author.