Treating Hypertension in Children With n-of-1 Trials

Participants were patients seen at the Pediatric Hypertension Clinics at McGovern Medical School, University of Texas Health Science Center in Houston, Texas, between June 2013 and July 2016. We included children (1) with primary hypertension and ≥9 years old, (2) with office hypertension (blood pressure [BP] exceeding the 95th percentile for age, height, and sex) over 3 separate visits, and (3) for whom the usual treating physician recommended treatment with an antihypertensive medication. We did not include patients with hypertension for whom a particular medication class might be indicated (eg, renin-angiotensin-aldosterone system blockade in patients with diabetes mellitus, chronic kidney disease, or proteinuria).

Eligible participants were approached during a routine clinic visit by the principal investigator (J.P.S.) or coinvestigator (J.A.S.) and provided a verbal and written description of the usual treatment approach and the additional procedures involved in the n-of-1 trial. Participation was optional and voluntary. Written informed consent was not required because the protocol was classified by the Committee for the Protection of Human Subjects at McGovern Medical School, University of Texas Health Science Center, as a quality improvement activity intended to optimize the care of individual patients.¹¹ 

Exclusion criteria were (1) contraindication to any of the tested medications and (2) parental or physician refusal for any of the tested medications.

The n-of-1 trials were designed to measure the relative ambulatory BP reduction in individual children and the acceptability of the side effects with 3 medications: amlodipine, hydrochlorothiazide, and lisinopril.¹² These medications are commonly used, inexpensive, taken once daily, and represent distinct medication classes with differing mechanisms of action.

The first step in the n-of-1 trial was to verify that antihypertensive medication was needed by confirming hypertension with 24-hour ambulatory blood pressure monitoring (ABPM). Using American Heart Association criteria, ambulatory hypertension was characterized by mean awake or sleep systolic blood pressure (SBP) or diastolic BP greater than or equal to the height-sex referenced 95th percentile with corresponding BP load (percentage of readings above threshold) >25%.^13,14 Adult cutoffs were used for patients aged 18 and above, with mean awake BP ≥135/85 or mean sleep BP ≥120/70.¹⁵ If ABPM (off medication) had been performed within the preceding 12 months, it was not repeated. For all others, antihypertensive medication was stopped for 2 weeks and ABPM was repeated to confirm the need for ongoing therapy.

The 3 medications were prescribed in a randomized sequence according to a computer-generated randomization schedule. For the first cycle, each of the 3 medications was taken for a 2-week treatment period (Fig 1). Treatment response for each medication was measured with 24-hour ABPM and a side effect questionnaire during the final 24 hours of each 2-week treatment period. Treatment response was only measured at the end of the second week to allow for washout of the previous medication. Carryover effects were thereby minimized because the known half-life of each medication suggested that effects of the previous medication would be negligible by then.^16,–18 

The second cycle of testing was restricted to 2 medications, each tested again for 2-week treatment periods. To narrow the choice of medication from 3 to 2 drugs, we excluded a medication if it produced an unacceptable side effect. Acceptability of side effects was judged by the child and parent and assessed by the questionnaire described in the Assessment of Adverse Effects subsection below. If none of the medications produced unacceptable side effects, then the 2 medications that resulted in the greatest reductions in mean awake SBP were retested in the second cycle. Thus, the design was tailored to the individual by selectively retesting medications in subsequent cycles based on patient response.

All participants received an identical regimen of conventional low-end weight-based doses in the first cycle.¹⁹ See Supplemental Table 5. Normal ambulatory BP was defined as mean SBP and diastolic BP for awake and sleep periods less than the height-sex referenced 95th percentile or adult cutoffs. If normal ambulatory BP was not achieved with low-end dosing, the dose was increased systematically in the subsequent cycle(s) according to protocol, until the BP normalized or maximum weight-based dosing was reached. The n-of-1 trial was considered complete when the preferred therapy was identified.

The preferred therapy was defined a priori as that which produced normal ambulatory BP and the greatest average reduction in mean awake SBP without unacceptable side effects for 2 treatment periods.

If >1 medication met these criteria, the medication with a more favorable side effect profile was chosen as the preferred medication for that individual. If none of the medications resulted in normal ambulatory BP despite maximal doses, the n-of-1 trial outcome was considered uncontrolled on monotherapy.

A side effect questionnaire was designed specifically for the n-of-1 trial and piloted in English and Spanish in a separate group of adolescents treated in our clinic. See Supplemental Fig 6. Laboratory evaluation was performed 2 weeks after starting lisinopril or hydrochlorothiazide, according to usual local practice. Adverse events, defined as hypertensive crisis, hypotension, >20% increase in serum creatinine from baseline with estimated glomerular filtration rate <90 mL/min/1.73 m², hypokalemia, or hyperkalemia, would prompt immediate discontinuation of the medication.²⁰ Pill counts were performed at the end of each treatment period. Medication adherence was calculated by dividing the actual number of pills missing by the number of pills expected to be missing.

As in routine clinical practice, we made no attempt to alter the medications to make them identical and indistinguishable to the patients or parents. When comparing the BP reduction and side effects of each therapy with the family, the physician was blinded to the drug name by using a summary sheet of ABPM results and side effects, which referred to treatments as first, second, or third used.

The preplanned analysis for the current study specified the use of Bayesian modeling. Conventional, frequentist modeling evaluates the probability that the observed data (or more extreme) would be obtained, assuming that the null hypothesis is true, and provides no estimate regarding the probability that the alternative hypothesis is true.²¹ As emphasized by the American Statistical Association, P values are widely misinterpreted, and many hypotheses beside the null or alternative hypothesis may be equally or more consistent with the observed data.²² Partly for these reasons, Bayesian analyses are increasingly used.^23,–26 Bayesian statistics are used to address a fundamentally different question that is more clinically relevant: what is the probability that the intervention has an effect on the outcome (or conversely, the probability that it does not), given the data obtained in the trial and any previous evidence?²⁷ Bayesian analyses do not involve a specific widely used probability threshold for “statistical significance” (analogous to a P < .05) and leave this judgement to the reader by presenting the data in a way that is more intuitively interpretable by clinicians and patients.

Bayesian approaches provide a point estimate of treatment effect as well as a 95% credible interval (CrI), defined as the interval that has a 95% probability of containing the true parameter value and is the Bayesian counterpart of the frequentist confidence interval.

The primary outcome of preferred therapy had 4 possible outcomes for each individual (amlodipine, hydrochlorothiazide, lisinopril, or BP uncontrolled on monotherapy). The primary outcome was analyzed with a multinomial model. In the absence of reliable previous evidence about the relative effectiveness of these medications for the population studied, we used a neutral probability with a wide CrI that any of the 3 drugs would be preferred to the other (∼Dirichlet [α = .25, .25, .25, .25]; where α represents the vector of priors for each multinomial category). Multinomial logistic modeling was used to evaluate predictors of the primary outcome, including sex, BMI z-score, African American race, and baseline mean awake SBP. Multinomial logistic regression coefficients used weakly informative priors ( ∼Normal [μ = 0, σ = 5]). Multilevel linear modeling provided estimates of mean awake SBP reduction with each medication compared with the baseline, nontreatment BP, with random intercepts by patient, thereby accounting for repeated measures within patients. Extremely vague, neutral priors for coefficients (∼Normal [μ = 0, σ = 1000]) were used as well as level 1 and level 2 variances (∼Folded-t distribution [degrees of freedom = 3, μ = 0, σ = 1000]). The models are described in more detail in the Supplemental Information. Any potential effect of treatment order was controlled for in 2 ways; study design included randomized sequence of treatments, and the analysis included time as a covariate in the multilevel model used to evaluate the magnitude of BP reduction with each medication.

From June 2013 until July 2016, 55 patients met criteria, and 42 out of 55 patients (76%) agreed to the n-of-1 trial (Fig 2). Among the 30 patients taking antihypertensive medication before enrollment, 20 completed an updated baseline ABPM because we required a recent (within 12 months) ABPM off medication to confirm the need for pharmacologic therapy. As a result, 23% (7 out of 30) were found to be normotensive and therefore did not require further visits in the n-of-1 trial. Three patients did not complete the first treatment cycle. The remaining 32 patients were included in the subsequent analyses.

Median age was 14 years (range: 9–19 years), and 72% were overweight or obese (Table 1). ABPM was completed every 2 weeks, a median of 6 times per patient (range: 3–8 times per patient). These ambulatory monitoring results were generally of good quality, with 87% (179 out of 205) containing at least 1 reading for at least 18 out of 24 hours.

As hypothesized, no single medication was preferred for the great majority of patients (Fig 3). Lisinopril was the preferred medication for the largest proportion of patients (49%; 95% CrI: 32% to 69%). The preferred medication identified by n-of-1 trial was generally not the same as prescribed before the study (Fig 4). Among the 6 patients taking multiple antihypertensive medications at enrollment, 83% (5 out of 6) had adequate BP control on monotherapy after the n-of-1 trial.

The preferred medication was largely unpredictable. In multinomial logistic regression analyses, estimates of association with preferred therapy had extremely large CrIs for most predictor variables (sex, BMI z-score, and baseline mean awake SBP). However, there was some evidence of increased preference of lisinopril over other treatments among African American participants, as 67% (8 out of 12) selected lisinopril as the preferred therapy. Specifically, African American participants had a high probability of selecting lisinopril over amlodipine or hydrochlorothiazide (posterior probabilities 89% and 98%, respectively). See Supplemental Table 6 for odds ratios and CrIs.

Multilevel modeling was used to evaluate mean awake SBP as a function of treatment and time (Fig 5). The baseline ABPM across participants off medication had a mean awake SBP of 134.4 mm Hg (95% CrI: 130.7 to 138.1). The posterior probability of decreasing mean awake SBP relative to baseline was ≥99% for all 3 medications. There was only a modest difference between medications in the mean change from baseline across all patients (−5.5 mm Hg [95% CrI: −8.9 to −2.0] for amlodipine, −6.1 mm Hg [−9.3 to −2.9] for hydrochlorothiazide, and −8.8 [−12.1 to −5.5] for lisinopril).

The posterior probabilities that lisinopril confers greater mean decreases in BP compared with amlodipine and hydrochlorothiazide across all patients assessed in this study are >99% and 98%, respectively. Average BP decreases associated with lisinopril versus amlodipine and hydrochlorothiazide were −3.3 mm Hg (95% CrI: −5.6 to −1.0) and −2.7 mm Hg (95% CrI: −5.1 to −0.3) respectively. The probability that amlodipine confers a decrease in BP versus hydrochlorothiazide is 68% (−0.6 mm Hg, 95% CrI: −3.1 to 1.9).

Despite only modest overall differences between medications, individual participants differed considerably in the measured effects on their BP. In comparing the greatest awake SBP reduction with the smallest reduction from baseline within individuals, the median difference was 12 mm Hg, (interquartile range: 6–21 mm Hg).

Most patients (84%, 27 out of 32) reported at least 1 side effect at any point during the n-of-1 trial (Table 2). Unacceptable side effects were reported with the greatest frequency on hydrochlorothiazide (25%), compared with 16% on lisinopril and 13% on amlodipine. No patients experienced hypertensive crisis, hypotension, hyperkalemia, or >20% increase in serum creatinine from baseline, and 2 patients on hydrochlorothiazide experienced mild hypokalemia that resolved after discontinuation of the medication.

The median medication adherence rate as assessed by pill counts was 100% (interquartile range: 57%–100%), likely an overestimation because pill counts were not performed in 28% of visits.

In this study in which we used n-of-1 trials and ABPM, lisinopril met our criteria for the preferred medication for the largest proportion of children treated for hypertension (49%, 95% CrI: 32% to 69%). As hypothesized, no single medication was preferred for a great majority (≥80%) of participants. Within-patient variation in BP response was considerable because the best-performing medication decreased the BP by about 12 mm Hg more than the worst-performing medication. Moreover, the medication selected by the n-of-1 trial was, for most patients, different from the pretrial therapy. Some patients receiving medications pretrial were found not to be hypertensive off treatment, and most patients treated with multiple medications at enrollment had BP control with conventional doses of only a single medication identified by the n-of-1 trial. These results illustrate the potential value in using an n-of-1 trial with ABPM to select the antihypertensive medication for an individual patient.

BP was assessed by ABPM, and we previously demonstrated that the ABPM diagnostic classification differed from the office BP classification in nearly 40% of visits.³⁰ Repeated ABPM was well tolerated by children with hypertension; >85% of monitorings were returned with a sufficient number of readings. Use of ABPM contributes to the frequent difference between pretrial and posttrial medication because pretrial therapy was often based on office BP readings, as is clinical practice.

The most important strength of this study was the use of an underused type of randomized trial to address an important clinical question that cannot directly be answered with a traditional parallel-group randomized trial: what is the best medication for my patient? Patients and families generally accepted the n-of-1 trial as a sensible approach to reduce therapeutic uncertainty. The repeated visits and ABPMs were not a deterrent for participation, and most families were willing to invest the added time and effort to identify the preferred therapy for their child. The additional cost of this approach compared with usual care was not measured in this study but should be investigated in future studies and related to the benefits. Approaches to minimize costs and maximize effectiveness require evaluation. Costs may be contained by maximizing the remote collection of digital data (including ABPM and side effect surveys) to minimize clinic visits. Reducing the additional time required of the clinician by protocolizing decision points will make this a more practical approach to integrate into usual practice.

African American patients are thought to have a less-robust antihypertensive response to angiotensin-converting enzyme inhibition compared with other racial groups, attributed to lower plasma renin levels and increased salt sensitivity.^31,32 In our preplanned subgroup analysis, lisinopril, the angiotensin-converting enzyme inhibitor, was preferred over both amlodipine and hydrochlorothiazide in 67% of African American participants.

Whether the best antihypertensive medication for most African American children differs from that for other children remains to be determined in large clinical trials powered to assess subgroup differences. Nevertheless, n-of-1 trials may be used to help optimize the treatment of individual children, particularly African Americans who have a greater prevalence, earlier onset, and greater severity of hypertension and a higher incidence of stroke, congestive heart failure, and end-stage renal disease.³³ 

In no previous pediatric studies have researchers directly compared the tolerability of antihypertensive medication classes. Hydrochlorothiazide had the greatest number of reported side effects, with a higher proportion bothersome enough that patients would not want to take the medication again. Adherence rates and BP control in clinical practice might be improved by more systematically inquiring about side effects.

More research is needed to refine the methods and analyses of n-of-1 trials, including the length and number of treatment cycles needed to identify clinically important differences in hypertension treatment. A potential limitation of our findings is that all important effects might not be identified in 2 weeks. However, on the basis of the half-life of the medications, we judged 2-week treatment periods would provide meaningful results and make the n-of-1 trials more feasible and acceptable. The generalizability of this approach may be improved by leveraging digital and remote data collection and automatizing decision pathways to ultimately integrate n-of-1 trials into routine clinical practice.

We found that n-of-1 trials with repeated ambulatory BP monitoring can be used successfully in children and adolescents with hypertension to identify a preferred antihypertensive regimen. This individualized approach to antihypertensive medication selection holds potential value by involving patients in their own care and facilitating informed treatment decisions. In usual care, the choice of specific antihypertensive regimen is based on physician preference with little or no systematic assessment of treatment benefits and hazards. A randomized trial in which researchers compare usual care to routine use of n-of-1 trials with ABPM is needed to assess effects on treatment adherence, patient satisfaction, long-term BP control assessed with ABPM, and hypertensive target organ damage.

ABPM

ambulatory blood pressure monitoring

BP

blood pressure

CrI

credible interval

SBP

systolic blood pressure