Personalizing Medical Treatment Decisions: Integrating Meta-Analytic Treatment Comparisons with Patient-Specific Risks and Preferences

Abstract

Background

Network meta-analyses (NMAs) that compare treatments for a given condition allow physicians to identify which treatments have higher or lower probabilities of reducing the risks of disease complications or increasing the risks of treatment side effects. Translating these data into personalized treatment plans requires integration of NMA data with patient-specific pre-treatment risk estimates and preferences regarding treatment objectives and acceptable risks.

Methods

We introduce a modeling framework to probabilistically integrate data from NMAs with data on individualized patient risk estimates for disease outcomes, treatment preferences (such as willingness to incur greater side effects for increased life expectancy), and risk preference. We illustrate the modeling framework by creating personalized plans for antipsychotic drug treatment and evaluating their effectiveness and cost-effectiveness.

Results

Compared to treating all patients with the drug that yields the greatest QALYs on average (amisulpride), personalizing the selection of antipsychotic drugs for schizophrenia patients would be expected to yield 0.33 QALYs (95% credible interval [crI]: 0.30–0.37) per patient over the next 5 years at an incremental cost of $4,849/QALY gained (95% crI: dominant–$12,357), versus 0.29 and 0.04 QALYs per patient when only accounting for risks or preferences, respectively, but not both.

Limitations

The analysis uses a linear, additive utility function to reflect patient treatment preferences and does not consider potential variations in patient time discounting.

Conclusions

Our modeling framework rigorously computes what physicians normally have to do mentally. By integrating three key components of personalized medicine – evidence on efficacy, patient risks, and patient preferences – the modeling framework can provide personalized treatment decisions to improve patient health outcomes.

INTRODUCTION

Network meta-analyses (NMAs) generate data to compare multiple treatments for a given condition, allowing physicians to identify which treatments produce a higher or lower relative risk reduction for disease complications or risk increase for treatment side effects. In contrast to traditional meta-analyses, NMAs allow for simultaneous comparisons of multiple treatment alternatives. Recent NMAs have evaluated antipsychotic,¹ antidepressant,² type 2 diabetes,³ arthritis,⁴ and blood pressure⁵ drugs, among others.

Physicians face the challenge of translating such data into personalized treatment decisions. This requires consideration of a patient’s absolute risks for different disease complications, and the patient’s preferences regarding treatment options such as willingness to incur more costs or risk increased side effects in exchange for reduced probability of disease complications. As Broekhuizen et al. note, “further research could … be directed toward the integration of patient-specific clinical outcome measures; that is, introducing a patient-specific performance distribution that yields estimates of the performance of a specific treatment for a specific patient… Such a holistic view of patient-specific variation in both preferences and clinical outcomes would be a step toward combining the two current viewpoints on personalized medicine [personalization by ‘physiology’ and personalization by ‘preferences’]”.⁶

Past approaches to treatment selection have accounted for population-level variations in risks and preferences,^7–11 patient-specific variations in risks,^12,13 or patient-specific variations in preferences,^6,14,15 without considering the combination of variations in both patient-specific risks and preferences.

Here, we introduce a modeling framework to probabilistically integrate NMA comparisons of the relative risks of treatment outcomes with patient-specific variations in absolute risks of those outcomes and preferences around outcomes. Our approach allows us to personalize medical treatment decisions and estimate uncertainty around how strongly a given treatment may be favored over another, as well as enabling personalized cost-effectiveness analysis. To illustrate, we apply the modeling framework to selection of antipsychotic drugs for patients with schizophrenia.

Schizophrenia is a mental disorder manifesting in symptoms that may include delusions, hallucinations, and disorganized thinking.¹⁶ Prevalence in U.S. communities is estimated to be between 0.33% and 0.75%.^17,18 The Global Burden of Disease Project estimates that schizophrenia was the 15^th leading cause of disability internationally in 2016.¹⁹ Schizophrenia symptoms may be mitigated with the use of antipsychotic drugs. However, the choice of antipsychotic drugs depends on many complex factors. A recent NMA evaluated 15 different available drugs and considered outcomes including efficacy, weight gain, extrapyramidal side effects (motor function difficulties), prolactin increase (which causes problems with the reproductive system), rate-corrected QT interval (QTc) prolongation (heart rhythm disturbances), and sedation.¹ The NMA found that no one drug was likely to be the best choice on all dimensions; rather different drugs were better at reducing the risks of different outcomes. For example, the drug with the highest efficacy for reducing psychotic symptoms (clozapine) also had high risk of causing several side effects. Prescription of antipsychotic drugs is also controversial because second-generation antipsychotic drugs, which constitute 13 of the 15 drugs considered by the NMA, are expensive. Globally, an estimated $14.5 billion was spent on these drugs in 2014.²⁰ The authors of the NMA concluded that clinicians should “adapt the choice of antipsychotic drug to the needs of individual patients.” However, there is currently no rigorous framework for making this decision. We apply our modeling framework to address this gap.

METHODS: Generalized Modeling Framework

Overview

Using data from an NMA, the modeling framework simulates 10,000 identical versions of a patient on each treatment alternative and computes treatment scores that account for both estimated values and uncertainty in the probability of disease outcomes and patient preferences. To account for patient risks, the framework simulates events and then creates scale ranges to facilitate comparison of changes in different disease outcomes. To account for patient preferences, the framework includes steps for preference elicitation, creation of a utility function that translates preferences and outcomes onto a consistent numeric scale, incorporation of the preferences and outcomes into a personalized cost-effectiveness ratio, and risk adjustment (Figure 1).

Figure 1. Overview of modeling framework.

We simulate 10,000 identical versions of a patient on each treatment and compute discounted numbers of events (e.g., disease events and occurrence of severe side effects) in each case. From this simulation, for each outcome, we compute a scale range. We then elicit patient preferences and compute utility scores for each version of the patient on each treatment. Finally, we use these utilities to compute QALYs and ICERs, and then risk adjust.

Event simulation

Using discrete event simulation, 10,000 identical versions of a patient on each treatment alternative are simulated using data from an NMA over the time period for which data are available. At each time step, the framework computes baseline (pre-treatment) absolute risks of each disease outcome, adjusts for the relative risk of each outcome for each treatment alternative based on the NMA data, and probabilistically estimates the occurrence of each disease outcome over the simulated period. Machine learning models are used to relate patient features to the pre-treatment risk of each outcome. Machine learning is a type of artificial intelligence that uses computers to automatically learn from data without being explicitly programmed. Machine learning models have been trained and validated to estimate the probability of numerous disease outcomes using a wide variety of approaches.^21–23 If not already in the literature, models can be trained using data from the placebo/comparator arms of the NMA. We demonstrate this latter approach in our schizophrenia example where we train random forest models and use cross-validation to estimate out-of-sample performance.

NMAs typically report results as odds ratios (ORs) or standardized mean differences (SMDs). Importantly, NMAs report relative treatment effects, which must be integrated with an individual patient’s baseline risk of each outcome to calculate absolute treatment effects. For example, a patient with a low baseline risk of a condition may not experience a substantial absolute risk reduction from a given relative risk reduction, whereas a patient with a high baseline risk of a condition may experience a substantial absolute risk reduction from that same relative risk reduction.

To adjust baseline absolute risks for each treatment alternative, the framework converts odds ratios (ORs) of disease outcomes from the NMA of randomized trials to relative risks (RRs) using baseline prevalence data from published studies. To estimate each outcome probabilistically, a binomial function with probability equal to the adjusted absolute risk is used. Finally, discounted numbers of events for each version of the patient on each treatment are computed using a 3% discount rate. Thus, for each treatment, the discounted number of events is calculated for each of the 10,000 versions of the patient.

If the NMA reports standardized mean differences (SMDs) of treatments versus placebo for an outcome (e.g., an outcome such as body weight, rather than a binomial outcome that could be codified as an event), the SMDs are converted to mean differences using pooled standard deviations calculated from published studies. The corresponding patient baseline risk factor values are then modified by the mean differences to simulate the effects of the different treatments.

Scale ranges

For each outcome, 95% credible intervals (crIs) are computed for the identical versions of the patient on each treatment. The scale range is taken as the interval hull of these 95% crIs for each outcome. The interval hull of m intervals is the smallest interval that includes all m intervals. For example, if the intervals were [1, 3], [2, 5], and [4, 6], the interval hull would be [1, 6].

Preference elicitation

There are many methods for preference elicitation.²⁴ In general, it is unclear which methods are best in which situations.²⁴ Any method can be used in the framework, with no additional modification, provided it can generate preferences for outcomes on a scale from zero to one that sum to one. Even if a method generates preferences on a different scale, it can still be used if the preferences can be standardized onto this scale. Preference standardization is discussed in the utility scores section.

Two common methods for preference elicitation include swing weighting and discrete choice experiments (DCE).^25,26 For swing weighting, patient preferences can be elicited using sliders, a method that has been shown to be visually intuitive and easy for patients with varied numeracy skills to understand.^27–29 Using sliders, a patient allocates weights to swings across the scale ranges such that the most important is assigned one and the rest are relative on a scale of zero to one. For example, in a case with two outcomes, nausea and having to inject a medication, the patient might be asked the relative importance of having no nausea as opposed to daily nausea, versus no injection as opposed to daily injections. The patient might input the weights 1 and 0.5 for the swings, respectively, in response. The sliders are marked with ticks every 0.1 increment, and the framework assumes 95% crIs from the next lowest to the next highest tick. For DCE, a patient repeatedly selects which set of outcomes the patient would prefer from two different hypothetical options. This information can then be used to compute preference weights and associated crIs. For a comparison of these two approaches and discussion regarding when they should be used, see Tervonen et al.²⁵ In the context of a single patient, swing weighting may be preferable as DCE might require the patient to state preferences over a large number of hypothetical alternatives.

Utility scores

The framework assumes linear partial value functions and a linear, additive utility function over the scale ranges, although other utility functions can also be used. A partial value function gives the value over a scale range for a single outcome. These forms are the most common in the literature^6,9 and are based on two key assumptions. First, it is assumed that there are no significant nonlinearities in value over the scale ranges: for example, the value associated with a one unit change at the lower end of a scale range is approximately equal to that associated with a one unit change at the upper end of a scale range. Second, it is assumed that the disutility associated with two conditions equals the sum of their individual disutility values. The validity of these assumptions must be considered for each application, possibly through solicitation of expert opinion of patients and physicians. Of note, only linearity of the partial value functions over the scale ranges is required, which reflect probable ranges of outcomes. This is a much weaker assumption than assuming linearity over all values of all outcomes. In our schizophrenia example, we make these assumptions based on the expert opinion of the senior author.

If these assumptions are violated, an alternative utility function can be used in the framework, with no additional modification. The appropriate functional form depends on the specific application. For example, for the scale range toxin level, a piecewise partial value function may be appropriate in which the function is relatively constant below some physiologically determined threshold and then rapidly decreases as the toxin reaches unsafe levels. If the appropriate functional form is not known, a technique such as conjoint analysis can be used.³⁰

To compute utilities from the partial value functions, preference weights are drawn from distributions for each of the 10,000 identical versions of the patient. These values are then standardized so that the weights w_i sum to one over all outcomes i = 1, …, n. The partial value functions u_i (c_i), where c_i is the value of outcome i, are also standardized such that, for undesirable outcomes, the value of being at the upper end of a scale range is 0 and the value of being at the lower end of a scale range is 1. Equation (1) is used to compute utilities for each of the 10,000 versions of the patient on each of the treatments:

$$u(\mathbf{c},\mathbf{w})=\mathbf{\sum }_{i=1}^n{w}_i*u_i\left(c_i\right)$$ (1)

Here u is utility, c is the vector of outcomes, w is the vector of preference weights for swings over scale ranges, and u_i (c_i) are the partial value functions. Thus, for example, for undesirable outcomes, a patient at the upper end of all scale ranges would have utility of 0 while a patient at the lower end of all scale ranges would have utility of 1.

Personalized cost-effectiveness

The quality-adjusted life-years (QALYs) associated with moving across all scale ranges from the upper to the lower ends for each of the n outcomes, Q_UL,i, are summed using equation (2):

$$Q_{UL}=\mathbf{\sum }_{i=1}^n{Q}_{UL,i}$$ (2)

The number Q_UL is then multiplied by the utility u for a patient on a treatment (from equation (1)) to obtain QALYs relative to the patient at the upper end of all the scale ranges. However, this reference point is problematic because the patient at the upper end of all the scale ranges would have zero utility and thus zero QALYs, but may not be dead over the model period. Therefore, the offset between QALYs for the dead patient and the patient at the upper end of all the scale ranges, which we denote by Q_DU, is added to Q_ULu:

$$Q=Q_{UL}u+Q_{DU}$$ (3)

The quantity Q is QALYs experienced by the patient over the model period. To calculate the offset Q_DU, the framework calculates QALYs for the patient if completely healthy over the time period under consideration versus dead (Q_DH), and subtracts QALYs associated with moving across all scale ranges (Q_UL) and QALYs associated with moving from the lower end of all scale ranges to completely healthy (Q_LH):

$$Q_{DU}=Q_{DH}-Q_{UL}-Q_{LH}$$ (4)

Figure 2 illustrates these QALY components.

Figure 2. QALY components.

The total QALYs for a patient from death to full health, Q_DH, can be computed as the sum of three components: (1) Q_DU, the QALYs from death to the upper ends of the scale ranges, (2) Q_UL, the QALYs from the upper to the lower ends of the scale ranges, and (3) Q_LH, the QALYs from the lower ends of the scale ranges to full health. The scale ranges are assumed to be undesirable in that being at the upper ends results in worse health; upper and lower can be interchanged if this is not the case.

To calculate a cost-effectiveness ratio, the increase in cost is divided by QALYs gained, which is calculated as the difference in Q between two treatment alternatives, both estimated by equation (3). This enables the framework to convert personalized measures of utility to measures of cost-effectiveness, enabling comparison of disparate interventions. If cost-effectiveness of a population-level program is desired, incremental costs and QALYs for the population, respectively, can be summed before dividing.

Risk adjustment

Past methods for personalizing treatment decisions have generally not accounted for patient risk preferences.^6,9 However, consider a treatment that guarantees 10 years of life and one that offers a 50/50 chance of 0 or 20 years. Even though these treatments yield the same expected number of years of life, different patients may prefer one or the other.

Numerous methods can be used to characterize patient risk preference. The framework assumes an exponential utility function based on previous proposals:³¹

$$u_R(Q)=a-b*{(r)}^{-Q}$$ (5)

Here, u_R is risk-adjusted utility, Q is QALYs, r is risk odds (computed from a single gamble), and a and b are standardizing constants. Other utility functions can be used, as appropriate. One advantage of this approach is that a patient’s utility function can be fully defined based on the result of a single gamble. See Supplement text SA1 for additional details.

A simple, generic illustration of the modeling framework for the case of three drugs and two outcomes (disease events and serious adverse events) is presented in Supplement text SA2. This example is intended to make the modeling framework easier to understand; however, it is not necessary to read the example to understand the present study.

METHODS: Application of generalized modeling framework to selection of antipsychotic drugs for schizophrenia

We applied the modeling framework to selection of antipsychotic drugs for patients with schizophrenia (full details in Supplement text SA3.1). We considered the 15 different drugs that were evaluated against each other or against placebo in a recent NMA.¹ All analyses were programmed in R version 3.5.1,³² with input data and statistical code for replication and extension of our analysis published at https://sdr.stanford.edu concurrent with publication. To use our modeling framework, we had to first define variables specific to the problem.

Outcomes

We considered nine outcomes, including the six outcomes from the NMA (efficacy, weight gain, extrapyramidal side effects, prolactin increase, QTc prolongation, and sedation) and the additional outcomes of agranulocytosis (a possible adverse immunological effect of clozapine), prescription drug cost, and years of life lost. Agranulocytosis incidence in each of the first two years after initiation of clozapine was obtained from the literature³³ and assumed to be zero thereafter.³⁴ We obtained costs for schizophrenia drugs from various sources (Table S2), and costs for conditions associated with schizophrenia (Table S3) from a past study.³⁵ Years of life lost for a patient were calculated as compared to expected years of life for a healthy person of the same age and sex in the United States.³⁶ All outcomes except agranulocytosis were assumed to be irreversible based on the senior author’s experiences treating patients with schizophrenia and current knowledge of the natural history of the disease.³⁷ If patients develop agranulocytosis, it was assumed to only affect them in the year that they develop it as the condition is treatable.

Population sample

We generated a set of 1,000 patients demographically representative of the adult population of the United States with schizophrenia, each with their own risk factors (age, sex, race, Positive and Negative Syndrome Scale (PANSS) score, prior psychiatric hospitalizations, weight, prolactin level, and QTc interval), based on data from the Collaborative Psychiatric Epidemiology Surveys (CPES).³⁸ From CPES we obtained means, standard deviations, and a correlation matrix for patient features. We randomly generated patient features using a multivariate random sample from the correlated distributions using copulas.

Personalized risk estimates

We generated baseline mortality rates for hypothetical versions of the patients of the same age and sex in the United States without schizophrenia based on data from the World Health Organization.³⁶ We calculated mortality rates for these hypothetical patients with moderate schizophrenia using a mortality multiplier from the literature that accounts for common non-therapy-related causes of heightened mortality among patients with schizophrenia (e.g., pneumonia related to homelessness, interpersonal violence related to poverty, etc.).³⁹ We then calculated mortality rates for the simulated patients by assigning both of these health states PANSS scores, based on the literature,⁴⁰ and assuming that mortality scales linearly with PANSS score, based on the expert opinion of the senior author. We also assumed that prolonged QTc interval and agranulocytosis increase mortality, based on prior literature.^33,41

We developed models relating patient features to associated baseline absolute risks for each potential outcome using data from the placebo arms of trials cited by the NMA.¹ We used random forest machine learning models⁴² and leave-one-out cross validation to estimate out-of-sample performance (Table S4) measured by root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). Figures S5–S6 show the variable importance metrics and partial dependence plots for the random forest models.

Patient preferences

In practice, preferences of individual patients would be elicited as described in the preference elicitation subsection, above. Here, we generated preferences for the simulated patients by repeatedly sampling from distributions based on a previous discrete choice experiment⁴³ and assuming independence between preferences. Specifically, we constructed gamma distributions for preferences based on reported preference means and uncertainty estimates. We used gamma distributions as these preferences were on a scale from zero to positive infinity. After drawing sets of preferences, we standardized them so that they summed to one. For discrete variables, no further work was required. For continuous variables, we scaled the preferences by the outcome scale ranges for the patient divided by the changes in the outcomes over which preferences were solicited in the discrete choice experiment. This assumes that these preferences scale linearly with the outcomes. Of note, the changes in outcomes over which these preferences were solicited in the discrete choice experiment were relatively broad, so we rarely had to extrapolate outside these ranges. Thus, for each patient, we sampled preferences from distributions for each outcome and then standardized them so they summed to one.

Selection of antipsychotic drugs for schizophrenia

We used these variables and our modeling framework to personalize selection of antipsychotic drugs for schizophrenia. Briefly, we simulated 10,000 versions of each patient in our population sample on each treatment alternative to project potential outcomes and scale ranges. The event simulation was run for 5 years to reflect clinical decision timelines and duration of studies reported in the NMA.

We used patient preferences from a previous discrete choice experiment and equation (1) to compute utilities for each of the 10,000 versions of each patient on each treatment alternative. Utilities were converted to QALYs using equation (3). We next computed mean QALYs, which were used to determine the best treatment alternative for each patient. We also calculated total U.S. societal costs associated with each treatment plan. Outcomes, utilities, QALYs, costs were discounted over the duration of each patient’s life at a rate of 3% annually, following current cost-effectiveness guidelines.⁴⁴ In this example, we did not risk adjust QALYs in the base case analysis as we did not have risk preference information specific to schizophrenia. In supplemental analysis, we considered the effects of risk adjustment.

We then calculated the incremental cost-effectiveness ratio (ICER) of personalized treatment determined using our modeling framework versus prescribing the best treatment for the population as a whole to all patients, which was also determined using our modeling framework. We adopted a U.S. societal perspective for both costs and QALYs. We used our previously calculated QALYs to select drugs but did not use them to calculate the ICER as they were influenced by patient preferences. Instead, to calculate the ICER, we used separately calculated QALYs from simulation outcomes using societal QALY weights. Specifically, we summed incremental simulated costs and summed the incremental simulated QALYs for the sample population before dividing the two quantities to compute the ICER.

RESULTS of Applied Schizophrenia Example

For our demographically representative sample of adults in the U.S. with schizophrenia, we found that amisulpride was the best antipsychotic drug in terms of maximizing mean QALYs. However, the distributions for many drugs overlapped with that for amisulpride (Figure 3), suggesting that personalizing treatment based on patient-specific risks and preferences can increase overall QALYs.

Figure 3.

Projected QALYs gained (versus placebo) for patients on different antipsychotic drugs. A demographically representative population of 1,000 patients with schizophrenia was randomly generated. For each patient, we simulated 10,000 identical versions of the patient on each drug. We computed mean quality-adjusted life years (QALYs) gained for each patient on each of the drugs versus placebo. The boxplots show median, interquartile, and outlier mean QALYs gained for patients on each drug. The order of the bars from left to right corresponds to that of the drugs in the legend.

We project that 0.33 QALYs (95% crI: 0.30–0.37) would be gained per patient by using our modeling framework to personalize schizophrenia treatment as compared to prescribing amisulpride to all patients (Figure 4). This benefit is driven by variations in patient-specific risks and preferences, which cause different treatments to be preferred for different patients. Compared to amisulpride for all patients, total incremental net present costs per patient of the personalized treatment plan over 5 years are $2,524, and its ICER is approximately $4,849/QALY gained (95% crI: dominant–$12,357).

Figure 4.

Projected QALYs gained (versus placebo) for patients using different methods for selecting antipsychotic drugs. A demographically representative population of 1,000 patients with schizophrenia was randomly generated. For each patient, we simulated 10,000 identical versions of the patient on each drug. We considered three strategies including 1) personalized treatment using our modeling framework, 2) population treatment in which the best drug for the population as determined from our modeling framework (amisulpride) is given to all patients, and 3) treatment in which the best drug as determined by efficacy (clozapine) is given to all patients. The boxplots show median, interquartile, and outlier mean QALYs gained versus placebo for patients using each method. The order of the bars from left to right corresponds to that of the methods in the legend.

Tables S5 and S6 show mean baseline risks and preferences, respectively, of patients allocated to different treatment alternatives, and Figure S7 shows the percent of patients allocated to each treatment alternative. In the personalized treatment plan, patients who have high baseline PANSS scores and preferences for drug efficacy are generally given clozapine (~36% of patients), which is the most effective drug but has severe side effects. Patients with lower baseline PANSS scores and preferences for drug efficacy, and who are more susceptible to and have stronger preferences for avoiding side effects, are given amisulpride (~41% of patients), which is the second most effective drug and has less severe side effects. Patients with less common sets of risks and preferences are given a variety of three other drugs including aripiprazole, haloperidol, and lurasidone (~15% of patients). Finally, patients who are more susceptible to and have stronger preferences for avoiding side effects are not prescribed drugs (~8% of patients).

Both patient-specific risks and preferences influenced drug selection. When we considered only patient-specific risks (Figure 5) or only patient preferences (Figure 6), different drugs were preferred under different plausible sets of patient input values (Table S7). Accounting for patient-specific risks but not preferences led to a gain of 0.29 QALYs per patient compared to prescribing amisulpride for all patients, and accounting for patient-specific preferences but not risks led to a gain of 0.04 QALYs per patient. When both of these factors were considered when making treatment decisions, 0.33 QALYs were gained per patient.

Figure 5.

Influence of baseline patient risks on drug choice. Drug scores were computed for three sets of potential baseline patient risks including relatively low risk of disease events, base case, and relatively high risk of disease events. For each set of risks, a simulation of 1,000 identical patients was run while sampling from distributions reflecting uncertainty in drug effects, baseline risks, and patient preferences. The boxplots show median, interquartile, and outlier QALYs for each set of baseline risks. The order of the bars from left to right corresponds to that of the drugs in the legend.

Figure 6.

Influence of patient preferences on drug choice. Drug scores were computed for three potential sets of patient preferences including relatively low preference for avoiding psychotic symptoms, base case, and relatively high preference for avoiding psychotic symptoms. For each set of preferences, a simulation of 1,000 identical people was run while sampling from distributions reflecting uncertainty in drug effects, baseline risks, and patient preferences. The boxplots show median, interquartile, and outlier QALYs for each set of preferences. The order of the bars from left to right corresponds to that of the drugs in the legend.

We considered the effects of risk adjustment in supplemental analysis (Supplement text SA3.2). In general, risk adjustment had relatively little effect: 80% of patients (95% crI: 77%–82%) were allocated to the same drug with and without risk adjustment. With risk adjustment, as the uncertainty regarding amisulpride efficacy increased, clozapine became the preferred treatment under a plausible set of patient input values (Figure S8). However, the difference was relatively minor as both drugs have similar levels of uncertainty in their performances on the outcomes. Risk adjustment would be expected to have a greater effect when some drugs have substantially higher levels of uncertainty than other drugs.

DISCUSSION

We have presented a generalized modeling framework for personalized medical treatment decisions which probabilistically integrates data from NMA drug comparisons with patient-specific risk factors and preferences. We applied the modeling framework to personalized selection of antipsychotic drugs and showed that QALYs could be gained – cost-effectively – using personalized treatments generated using our modeling framework compared to using the same treatment for all patients. Our analysis underscores the importance of considering both patient-specific risks and preferences when making treatment decisions: when these factors were not considered, the resulting personalized treatments yielded worse health outcomes than when they were considered.

Our modeling framework fills an important gap in the personalized medicine literature: NMAs are increasingly common but frequently contain information on so many outcomes that it is difficult for physicians to integrate all the data for a particular patient. When available, absolute risk calculators are helpful for calculating pre-treatment risk, but still require physicians to calculate relative risk reductions and integrate them with patient preferences. Our modeling framework allows for rigorous integration of evidence on efficacy, patient risks, and patient preferences.

Our modeling framework has several limitations. First, we use linear partial value functions and a linear, additive utility function; thus, we assume that there are no significant nonlinearities in value over the scale ranges and that the disutility associated with two conditions is equal to their sum. These assumptions are the most common in the literature.^6,9 However, it is important to be aware of them in each application of the framework and use alternative forms if needed. Second, we suggest the use of sliders to solicit patient preferences (although we did not use them in our analysis of antipsychotic drugs). Different methods for preference solicitation are currently being evaluated.^27–29,31 Our modeling framework can incorporate another method if found to be preferable. Third, we do not include potential variations in patient time discounting. Finally, based on the literature and ease of clinical use, we risk adjust scores using exponential utility functions, which require that patient preferences approximately satisfy the delta property within the range of payoffs (Supplement text SA1). Other methods for risk adjustment can be used,³¹ depending on application and user preference. The applied example of our modeling framework to antipsychotic drug selection has the additional limitation that we assume mortality scales linearly with PANSS score based on expert opinion of the senior author.

Future work could test the use of our modeling framework for antipsychotic drug selection in clinical settings. To do so, it might be helpful to perform a randomized pilot study comparing physicians and patients using our approach versus using standard care approaches, and identify changes in key outcome measures such as the decisional conflict scale,⁴⁵ which indicates satisfaction with a medical decision, and changes in longer-term outcomes from treatment. In future work, our modeling framework could also be applied to personalize drug treatment for other diseases for which multiple treatments, each with different risks and benefits, produce complex clinical decision-making dilemmas for physicians and patients.

Our modeling framework rigorously computes what physicians normally have to do mentally and integrates three key components of personalized medicine: evidence on efficacy, patient risks, and patient preferences. Personalizing medical treatment decisions through consideration of both patient-specific risk estimates and preferences can improve both patient satisfaction and health outcomes.