Article Text

Original research
Interpreting hydroxychloroquine blood levels for medication non-adherence: a pharmacokinetic study
  1. Stephen Balevic1,2,3,
  2. Kai Sun3,
  3. Jennifer L Rogers3,
  4. Amanda Eudy3,
  5. Rebecca Eli Sadun3,
  6. Mithu Maheswaranathan3,
  7. Jayanth Doss3,
  8. Lisa Criscione-Schreiber3,
  9. Tyler O'Malley4,
  10. Megan Clowse3 and
  11. Daniel Weiner5
  1. 1Department of Pediatrics, Duke University School of Medicine, Durham, North Carolina, USA
  2. 2Duke Clinical Research Institute, Durham, North Carolina, USA
  3. 3Department of Medicine, Duke University School of Medicine, Durham, North Carolina, USA
  4. 4Exagen, Vista, California, USA
  5. 5UNC Eshelman School of Pharmacy, Chapel Hill, North Carolina, USA
  1. Correspondence to Dr Stephen Balevic; stephen.balevic{at}


Objective Characterise the relationship between hydroxychloroquine (HCQ) blood levels and the number of missed doses, accounting for dosage, dose timing and the large variability in pharmacokinetics (PK) between patients.

Methods We externally validated a published PK model and then conducted dosing simulations. We developed a virtual population of 1000 patients for each dosage across a range of body weights and PK variability. Using the model, 10 Monte Carlo simulations for each patient were conducted to derive predicted whole blood concentrations every hour over 24 hours (240 000 HCQ levels at steady state). To determine the impact of missed doses on levels, we randomly deleted a fixed proportion of doses.

Results For patients receiving HCQ 400 mg daily, simulated random blood levels <200 ng/mL were exceedingly uncommon in fully adherent patients (<0.1%). In comparison, with 80% of doses missed, approximately 60% of concentrations were <200 ng/mL. However, this cut-off was highly insensitive and would miss many instances of severe non-adherence. Average levels quickly dropped to <200 ng/mL after 2–4 days of missed doses. Additionally, mean levels decreased by 29.9% between peak and trough measurements.

Conclusions We propose an algorithm to optimally interpret HCQ blood levels and approximate the number of missed doses, incorporating the impact of dosage, dose timing and pharmacokinetic variability. No single cut-off has adequate combinations of both sensitivity and specificity, and cut-offs are dependent on the degree of targeted non-adherence. Future studies should measure trough concentrations to better identify target HCQ levels for non-adherence and efficacy.

  • Pharmacokinetics
  • Antirheumatic Agents
  • Lupus Erythematosus, Systemic

Data availability statement

Data are available upon reasonable request. The pharmacokinetic model used for dosing simulations is publicly available from the original publication.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Hydroxychloroquine (HCQ) whole blood levels are commonly measured in clinical practice and research studies to determine medication non-adherence.

  • Prior studies have not been able to quantify the number of missed doses that correspond with HCQ blood levels, accounting for all of the confounders known to impact drug level measurements.


  • No single HCQ level cut-off has adequate combinations of both sensitivity and specificity, and cut-offs are dependent on the degree of targeted non-adherence.

  • Simulated HCQ levels quickly dropped to <200 ng/mL after 2–4 days of missed doses, and mean levels decreased by 29.9% between peak and trough measurements.


  • An algorithm was developed to optimally interpret HCQ blood levels and quantify the number of missed doses, incorporating the impact of dosage, dose timing and the wide pharmacokinetic variability between patients.

  • Future studies should measure tough concentrations to better identify target HCQ levels for efficacy.


For patients with systemic lupus erythematosus (SLE), regularly adhering to medications such as hydroxychloroquine (HCQ) can be challenging. Depending on the measure used, 43%–75% of patients with SLE exhibit some degree of medication non-adherence.1 Identifying medication non-adherence is crucial for patient care, as patients with inadequate exposure to HCQ may be 2.5× more likely to have a severe lupus flare and accrue more damage.2 3 Furthermore, medication non-adherence is modifiable, with studies showing that counselling patients can improve the number of doses taken.4 5 Despite the importance of identifying medication non-adherence in SLE, measures to predict medication usage differ in their sensitivity and specificity, and adherence is not static and may change over time,6 making it difficult to accurately detect.

Recently, measuring HCQ blood levels as a surrogate to identify medication non-adherence or obtain ‘therapeutic’ targets has received great attention in the rheumatology community.7 A meta-analysis found seven studies that compared HCQ blood levels to patient or physician reported adherence, or pharmacy refill records, and found three times higher odds of reported non-adherence in patients with low HCQ levels.8 Various cut-offs for non-adherence have been proposed for HCQ whole blood levels, with most ranging from 100 ng/mL to 500 ng/mL.9–13 However, these prior studies have not been able to account for all the factors known to impact drug level measurements, including the dose amount, the time since the dosage was last taken, interindividual variability in drug pharmacokinetics (PK) and the precise number and duration of time the doses were taken. Due to the poor correlation between HCQ weight-based dosing and drug levels,14 as well as the large interindividual variability in drug PK,15 a single blood level cut-off to determine medication adherence is likely to misclassify certain patients.

To address this knowledge gap, we leveraged a published population PK model for HCQ to conduct dosing simulations with the goal of (1) relating the number of missed doses to HCQ blood levels; (2) determining how quickly HCQ blood concentrations fall once the medication is discontinued and (3) characterising the impact of HCQ PK variability, dosage and timing since last administration on blood level measurements.

Materials and methods

This was a population PK modelling and simulation study. Data from human subjects were only used to externally evaluate the performance and suitability of the published PK model. Briefly, we implemented the population PK model in whole blood developed by Morita et al in Japanese adults receiving HCQ for SLE or cutaneous lupus erythematosus.16 This was a 1-compartment structural model with a covariate effect of body weight on apparent HCQ clearance (CL/F), estimation of apparent volume of distribution (V/F) and interindividual variability for clearance (ω2=0.0948). Based on the publication, weight was assumed to be modelled using a non-linear equation, and intraindividual variability was assumed to be modelled using an exponential error model.16 Using the following PK model equations and estimates from Morita et al, we simulated individual predicted concentrations from the model:

Embedded Image

Embedded Image

Embedded Image

Embedded Image

All model simulations were conducted in Phoenix NLME (Certara, Princeton, New Jersey, USA, V8.4) and data manipulation was conducted using RStudio V.1.4.1717 (RStudio, Boston, Massachusetts, USA).

Data for external validation were obtained from the Duke Lupus Registry (DLR). Whole blood concentrations for HCQ in the DLR were measured using liquid chromatography with tandem mass spectrometry (LC/MS/MS) by Exagen; the lower limit of quantitation was 20 ng/mL, inter-assay precision (%coefficient of variation (CV)) was 3.2–5.4, and intraassay precision (%CV) was 2.7–6.5 across the analytical measuring range.

External PK model validation

Because the population PK model by Morita et al was developed in a population of Japanese adults with potentially different covariates (eg, disease phenotype, weight, renal function, genotypes) than in the US population with SLE, we first sought to externally evaluate this model using random whole blood HCQ levels obtained from the DLR. The methods employed for the external model validation are outlined in online supplemental materials.

Supplemental material

Dosing simulations

For each dosing group, we conducted Monte Carlo simulations consisting of 1000 virtual patients, with blood sampling every hour over a 24-hour dosing interval at approximately steady state (eg, at least 3 months after first dose), and 10 simulation replicates per virtual patient, resulting in a total of 240 000 simulated concentrations. These simulations incorporated the PK model’s between-patient variability in HCQ clearance and body weights derived from the minimum and maximum observed weights for each dosage in the Morita study (online supplemental materials). In a subgroup analysis using only trough measurements, we conducted 100 simulation replicates (100 000 concentrations). For dosing, we simulated a once daily dose of HCQ sulfate (as the salt) every 24 hours for 100 doses per patient, representing 100 000 doses across the entire virtual population. Because 300 mg HCQ tablets are now commercially available for patients, we also simulated the 300 mg group as a once daily dosage (eg, we did not simulate alternating doses of 400 mg and 200 mg as is sometimes done in clinical practice).

To determine the impact of missed doses on simulated HCQ blood levels, we deleted at random a fixed proportion of doses across the entire 1000 virtual patients (100 000 doses) using the delete MCAR function (missMethods package in R). By deleting doses at random, we sought to mimic the variability in when patients in the general population might miss a dosage in a real-world setting. We evaluated an average proportion of the following missed doses: 20%, 40%, 60% and 80%; these correspond to the population missing an average of 20/100, 40/100, 60/100 and 80/100 doses over the course of approximately 100 days.


External validation of the Morita PK model

As outlined in online supplemental materials, there were a total of 137 random HCQ concentrations in the 400 mg DLR validation group; 15 in the 300 mg group and 33 in the 200 mg group. Demographics and clinical characteristics of the cohorts are noted in online supplemental table 1. Results from the validation exercises are noted in online supplemental figures 1 and 2.

The validation exercises demonstrated that the Morita model has good predictive performance for the DLR population and was suitable for the modelling and simulation methods used in the remainder of this manuscript.

Impact of chronic missed doses on HCQ blood levels

Average HCQ concentrations progressively declined as the proportion of missed doses increased (figure 1) and the number of concentrations falling below certain thresholds was highly dependent both on dosage and degree of non-adherence. For example, across all simulations where 40% of doses were missed, random concentrations <200 ng/mL (eg, levels throughout the entire dosing interval) occurred approximately 30% of the time at the 200 mg dosage, but less than 10% of the time at the 400 mg dosage.

Figure 1

Mean HCQ blood level and adherence. Colours represent degree of missed dosages. Dotted black line represents 200 ng/mL, a commonly used cut-off for non-adherence,11 as a reference. HCQ, hydroxychloroquine.

For patients receiving HCQ 400 mg once daily, random whole blood concentrations <200 ng/mL were exceedingly uncommon in those who were fully adherent (<0.1%), suggesting PK variability alone is unlikely to account for very low HCQ concentrations (table 1). However, when 80% of doses in the population are missed, approximately 60% of simulated concentrations were <200 ng/mL and 44.6% were <100 ng/mL. The number of low blood HCQ concentrations for the 200 mg and 300 dosages is noted in online supplemental tables 2 and 3.

Table 1

Percent of HCQ blood concentrations by degree of non-adherence (400 mg)

Impact of recent discontinuation on HCQ blood levels

As noted in figure 2, HCQ blood levels were impacted by recent medication discontinuation. For fully adherent patients receiving 400 mg daily, the average HCQ blood level remained above 100 ng/mL for approximately 6.2 days after the last dose was taken and above 200 ng/mL for 4.3 days. For fully adherent patients receiving 300 mg and 200 mg dosages, average concentrations remained above 100 ng/mL (200 ng/mL) for approximately 5.5 (3.7) and 4.2 (2.5) days, respectively.

Figure 2

HCQ blood levels after discontinuation. Dotted lines represent 100 and 200 ng/mL for reference. HCQ, hydroxychloroquine.

Impact of sampling peak versus trough levels

For fully adherent patients receiving 400 mg daily, the mean HCQ concentrations decreased by approximately 29.9% between peak (4 hours after dose) and trough (24 hours after dose) measurements, with mean concentrations of 1123 and 787 ng/mL, respectively.

Sensitivity and specificity of HCQ blood levels for non-adherence and proposed algorithm

When excluding samples arising from recent medication non-adherence (>4 days for the 400 mg dosage, >3 days for the 300 mg dosage and >2 days for 200 mg), figure 3 demonstrates the impact of using two hypothetical cut-offs (200 and 500 ng/mL) for classifying non-adherence using random blood levels.4 11 Results using only trough levels are noted in online supplemental figure 3.

Figure 3

Hypothetical random HCQ level cut-offs and adherence categories. Circles represent mean HCQ concentration across all time points, error bars represent 1 SD. Excludes samples >96 hours for 400 mg; >72 hours for 300 mg and >48 hours for 200 mg. HCQ, hydroxychloroquine.

For patients receiving 400 mg, a cut-off of <200 ng/mL would be expected to classify concentrations only arising from severe medication non-adherence with few to none from fully adherent patients (ie, high specificity). However, this cut-off would miss the majority of all instances where 80% of dosages were missed, as well as less severe non-adherence (ie, low sensitivity). Conversely, a <500 ng/mL cut-off would classify the majority of samples arising from severe non-adherence but would also exclude some with minimal non-adherence (ie, high sensitivity, low specificity).

Due to the wide variability in HCQ blood levels, there did not appear to be a single cut-off with an acceptable sensitivity and specificity for any dosage. To best account for the impact of dosage and time since the last HCQ intake, a proposed algorithm for monitoring HCQ blood levels and adherence is noted in figure 4.

Figure 4

Proposed algorithm to interpret HCQ levels and adherence. *Acknowledge up to 30% variation in levels between peak and trough. **Figures depict the majority of expected concentrations (mean±SD) from the population, excluding recent non-adherence. Individualise interpretation as described in the manuscript. HCQ, hydroxychloroquine.


We conducted this PK modelling and simulation study to clarify common misconceptions regarding drug-level monitoring for medication non-adherence and provide practical guidance for rheumatologists who are measuring HCQ blood concentrations as part of their clinical or research practice. Our results highlight that while HCQ blood levels can be a powerful tool to identify medication non-adherence, the principles outlined in this study are important for practitioners and researchers to avoid misclassifying patients.

Medication non-adherence exists along a spectrum; with some patients not taking any doses, some taking all expected dosages, and the majority somewhere in between.6 Due to the wide variability in HCQ blood levels between patients, we found that a single cut-off will invariably be prone to misclassifying adherence for at least some patients. Moreover, ‘drug forgiveness’ (the degree of non-adherence that results in loss of clinical efficacy) for HCQ is unknown. That said, we found that across 240 000 simulated, random whole blood levels from 1000 virtual patients, very low HCQ blood levels due only to PK differences are quite rare. For example, in completely adherent patients receiving 400 mg daily, only 0.03% of simulated blood levels were <200 ng/mL. Similarly, in completely adherent patients receiving 200 mg daily, approximately 2% were <200 ng/mL. Accordingly, whole blood levels <200 ng/mL are strongly suggestive of some degree of medication non-adherence regardless of dose, with progressively lower levels having higher specificity for, and higher degrees of, missed doses. Conversely, levels 300–500 ng/mL are less specific and can occur due to PK variability alone in fully adherent patients, particularly those receiving less than 400 mg daily. Therefore, we highlight that the sensitivity/specificity of any HCQ level cut-off is dependent on dosage (and dose time), as well as the degree of targeted non-adherence.

Although HCQ has a long-terminal elimination half-life of approximately 40 days, concentrations decline quicker than most rheumatologists expect once the medication is discontinued or dosages are missed. This is in part due to the fact that decline in concentrations is multiphasic; representing distribution into tissues, elimination by the kidneys and hepatic enzymes, and slow release of tissue-bound HCQ back into the blood.17 In addition, concentrations may decline quickly in the setting of non-adherence because patients may not be at steady state. Depending on dosage and steady state, the model predicted that average blood levels could decline to <100 ng/mL after approximately 4–7 days following the last intake. Conversely, another study suggests HCQ concentrations can remain high for up to 16 days after discontinuation although interpretation is limited by their sampling windows, dosing and statistical methods.18 Additionally, we found the within-day variation of HCQ over a 24-hour period is approximately 29.9%; which is highly consistent with other PK studies reporting a change of 27%.19 Therefore, it is essential that rheumatologists ask patients when they took their last HCQ dosage prior to measuring blood levels in order to distinguish ‘chronic’ or ‘regular’ medication non-adherence from ‘recent’ medication non-adherence. It would seem to reason that measuring trough HCQ concentrations may increase the sensitivity of non-adherent HCQ levels whereas measuring peak concentrations may increase the specificity.

Several studies have investigated the relationship between HCQ drug levels and clinical or safety outcomes, with reported ‘therapeutic’ targets in whole blood ranging between 500 and 1000 ng/mL.8 20 21 However, virtually all studies relied on random drug concentrations. Our modelling suggests, consistent with other literature,19 that individual patients could be incorrectly classified as being therapeutic versus subtherapeutic due simply to the timing the blood sample was drawn (eg, peak or trough levels). Accordingly, target HCQ blood levels from the literature relying on random drug concentrations should be interpreted with caution, and future research studies should measure trough concentrations to reduce variability and more accurately relate levels with clinical outcomes and identify optimal therapeutic targets. Additionally, our results cannot be used to inform optimal dosing strategies for patients at this time. However, once optimal HCQ blood levels are identified, it is likely that the current HCQ dosing paradigm may shift away from weight-based dosing and towards individualised dosing based on therapeutic drug level monitoring.

Although our results suggest that HCQ drug concentrations have an initial rapid decline once a dosage is discontinued, tissue concentrations are higher than that of blood. Additionally, it is likely that HCQ’s (multiple) pharmacodynamic effects on immune cells may persist for some time; collectively this results in a delay between medication discontinuation and loss of clinical efficacy, although this may also be dependent on the patient’s adherence prior to discontinuation.2 Conversely, the impact of using a ‘loading dose’ on clinical outcomes requires further investigation; one clinical trial in RA did report improvement with loading doses of HCQ, though the dose–response relationship was inconsistent.22

While several studies have investigated the relationship between HCQ blood levels and medication adherence, our study has several notable strengths and innovations. First, PK modelling and simulation allowed us to evaluate each variable in isolation; that is, we were able to distinguish the effect of timing since the last HCQ administration versus the number of missed dosages. Second, we could uniquely incorporate the impact of different dosages, sampling times and the wide variability between patients in HCQ PK across a variety of body sizes. Additionally, our simulated concentrations align well with observed concentrations in an external cohort, as well as other published population PK models23 and physiologically based PK models24 for HCQ, suggesting the principles outlined in this manuscript are generalisable to other populations.

Despite the study’s strengths, there are also limitations to the study. First, the simulations were derived from a published population PK model and therefore represent predicted, not observed, HCQ concentrations. All PK models have inherent limitations; and there are three important considerations with the Morita PK model, including (1) no covariate effect for weight on volume of distribution, which could result in overprediction of peak concentrations or underprediction of half-life for obese patients; (2) a narrow patient population with a relatively lower median body weight and unknown renal function and (3) use of a 1-compartment model, which may not fully characterise different phases of HCQ elimination. As a result of these limitations, we could be overestimating the rapid decline in HCQ concentrations after discontinuation, and it is also possible that the PK model will not predict concentrations well in patients who have extreme obesity or renal impairment, for which observed concentrations might be higher. Despite this potential limitation, we conducted external validation exercises and found that the model has overall good predictive performance. Residual differences in model predicted versus observed concentrations are likely multifactorial, due to differences in dose timing, adherence, potential misclassification in dose and individual patient differences in HCQ metabolism or other pharmacokinetic processes. It is also worth noting there are important differences between the Morita study and our validation cohort, including different numbers of male patients, although Morita did not find sex influenced HCQ PK. Additionally, we considered steady state to occur after approximately 3 months of treatment based on when blood levels plateau; however, some studies suggest approximately 6 months of HCQ treatment are needed to achieve steady state based on the terminal elimination half-life.25 This is unlikely to impact our results, as our simulations achieved steady state based on the Morita PK model. Lastly, for simplicity, we did not conduct simulations using HCQ dosed more than once a day (eg, two times per day dosing) or include residual variability in the PK model. Taken altogether, our simulations are best viewed as providing a key conceptual framework for clinicians interpreting HCQ blood levels.

In summary, our data suggest a stepwise approach to appropriately interpret HCQ blood levels for non-adherence. First, rheumatologists should ascertain when patients last took their HCQ dosage, which should ideally have occurred within the last 24 hours, but no later than 4–7 days prior (for the 400 mg dosage). Next, the figures and tables in this manuscript can be used to approximate the degree of non-adherence at the population level based on dosage and desired sensitivity/specificity. For example, at the 400 mg dosage and a threshold of 60% missed doses, approximately 70% of random levels are expected to be <500 ng/mL. Conversely, about 6% of random samples from fully adherent patients receiving 400 mg are expected to be <500 ng/mL. Lastly, rheumatologists can reduce the variability in HCQ blood levels by up to 30% by standardising sample collection around the same time after last dosage (eg, trough measurements), realising this is more logistically difficult in a clinical setting. In addition, practitioners must remain aware of reasons an individual patient’s blood levels may deviate from expected, including gastrointestinal conditions leading to impaired absorption and lower blood levels,26 significant renal dysfunction leading to reduced clearance and higher blood levels,27 extremes of body weight,14 28 pregnancy15 and drug–drug interactions.


We propose a stepwise approach to optimally interpret HCQ blood levels and estimate non-adherence that considers the impact of dosage, timing of last dose administration and the wide pharmacokinetic variability between patients. Non-adherence exists on a spectrum; with progressively lower HCQ concentrations having higher specificity for, and degrees of, medication non-adherence. For the average patient receiving HCQ 400 mg daily, blood levels <200 ng/mL are best explained by a non-adherent patient discontinuation of the drug >4 days prior to sampling, or by the patient missing ≥80% of dosages on a regular basis. However, interpretation of blood levels at an individual level is complex and no single cut-off has an adequate combination of both sensitivity and specificity. Future studies should measure tough concentrations to better identify target HCQ levels for efficacy.

Data availability statement

Data are available upon reasonable request. The pharmacokinetic model used for dosing simulations is publicly available from the original publication.

Ethics statements

Patient consent for publication

Ethics approval

Data for external validation were obtained from the Duke Lupus Registry (DLR). Patients consented to be in the DLR (Pro00008875); the study was approved by Duke Health IRB (Pro00094645) and was conducted in compliance with the Helsinki Declaration. Participants gave informed consent to participate in the study before taking part.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • X @MithuRheum

  • Contributors SB conceptualised/designed the study, analysed data and drafted the manuscript. KS, JLR, AE, RES, MM, JD, LC-S and MC contributed to data acquisition and interpretation. TO contributed to data interpretation. DW contributed to the study design, data analysis and interpretation. SB acts as the guarantor and accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish. All authors critically revised and approved the final manuscript.

  • Funding Support for this study was provided by the NIAMS (5K23AR075874).

  • Competing interests SB receives support from the National Institutes of Health, the Childhood Arthritis and Rheumatology Research Alliance, consulting for UCB and Rutgers University and serves on an NIH DSMB. KS is supported by the National Center for Advancing Translational Sciences of the National Institutes of Health, the American Heart Association COVID-19 Fund to Retain Clinical Scientists Award and the Duke REACH Equity Career Development Award. JLR has received grant support from Pfizer, Exagen, Immunovant, Astra-Zeneca and consulting fees from GlaxoSmithKline, Amgen, Aurinia, Immunovant, Janssen, Eli Lily and Ampel Biosolutions. AE has received grant support from Pfizer, Exagen, Immunovant and GlaxoSmithKline and consulting fees from Amgen. RES receives grant support from the Arthritis Foundation, the Childhood Arthritis and Rheumatology Research Alliance and the Rheumatology Research Foundation. MM is a consultant for Astra Zeneca. LC-S receives grant support through UCB and the arthritis foundation. TO is an employee of Exagen. MC has received grant support from Pfizer, Exagen, Immunovant and Astra-Zeneca and consulting fees from GlaxoSmithKline, Amgen and UCB. DW is an independent director for Simulations Plus. HCQ Blood concentrations for the validation cohort were measured and paid for by Exagen Diagnostics.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.