Development of a Conceptual Model of Childhood Asthma to Inform Asthma Prevention Policies

Background

Asthma remains a major cause of morbidity and economic burden across the globe. Asthma is the leading cause of absence from school and the third leading cause of productivity loss.^1–3 We have estimated the 20-year burden of uncontrolled asthma at $960B in the US, with more than $300B in direct costs alone.⁴ There is no cure for asthma and underdiagnosis in children might lead to permanent lung damage due to tissue remodeling⁵.

The Canadian Healthy Infant Longitudinal Development Study (CHILD)⁶, an ongoing multiethnic birth cohort of 3,455 families launched in 2008, provides an unprecedented opportunity to explore biological and environmental factors that can potentially be modified in early life to prevent asthma. The CHILD team and other investigators have identified antibiotic use, formula feeding, and exposure to phthalates as three modifiable early-life risk factors that can predict asthma diagnosis at the age of five.^7–14 Emerging evidence suggests a causal relationship between these factors, one that is potentially mediated through either microbiome, immune system, or possibly both.^7–10

Translation of such knowledge into practice, be it guideline recommendations for risk factor modification or policy-level decisions such as rolling out screening programs, will require careful evaluation of the value of any proposed intervention, which can be quantified in terms of the amount of resources that such interventions consume to produce one unit of health gain. Computer models are an indispensable part of such projections, as long-term outcomes of preventive strategies depend on a tangled network of clinical, economic, and behavioral decisions and factors. These models collect evidence from multiple sources and enable projection of intermediate outcomes (e.g., to what extent an antibiotic stewardship program will reduce the risk of asthma) to policy-relevant metrics (e.g., costs and Quality-Adjusted Life Years). In doing so they enable extrapolation beyond the typically short follow-up period of empirical studies to the longer time horizon that is required for properly capturing all the relevant consequences of a policy decision. For these reasons, the use of computer models for policymaking in health is considered “an unavoidable fact of life”.¹⁵

Conventionally, researchers develop a new decision-analytic model for each specific policy question at hand. However, this fragmented approach is inefficient and leads to unnecessary repetition of model development by many research groups, often with inconsistent assumptions.¹⁶ This inconsistency makes it hard to compare the results of similar studies. Lack of transparency that comes with insufficient documentation of such de novo models makes scientific replication difficult, and ultimately affects the credibility of projections.¹⁷ Our systematic review of decision-analytic models for asthma interventions concluded that currently-available models lack the granularity needed to inform policy on personalized, risk-prediction-based strategies.¹⁸ An alternative to such de novo modeling approach is to decouple the model development process from the policy question, thus creating ‘reference models’ that can be used to address multiple policy questions within a unified framework.¹⁶ This not only increases the consistency across different policy decisions but also enables proper consideration of ‘interactions’ among these decisions. For example, large-scale risk-factor modification attempts for asthma (e.g., reduction in exposure to antibiotics) will change the prevalence of asthma, which in turn affects the yield and cost-effectiveness of pediatric asthma screening programs. Reference models can account for these complex interactions.

As the CHILD study and other global initiatives identify more preventive targets for childhood asthma, we recognize the need for a reference decision-analytic model to evaluate long-term effects of these prevention strategies. Here, we review models focused on the primary prevention of asthma and report the formal process through which an expert panel of CHILD study investigators discussed and drafted the concept map for an asthma policy model that reflects the current state of our knowledge about asthma diagnosis.

Methods

The ISPOR-SMDM Modeling Good Research Practices Task Force has advocated for an explicit process of consulting experts and engaging with stakeholders to convert the problem into the appropriate model structure using influence diagrams and concept maps.¹⁹ We followed these steps, starting from a scoping review of decision-analytic models of interventions related to primary prevention of asthma to identify any potential models that can be adapted for evaluating asthma prevention strategies in Canada. This was followed by a formal process of consensus building on both the decision problems that the model will have to tackle and the conceptual structure of such a model in terms of the important domains and associations that need to be considered. The final outcome of this exercise was a concept map of an asthma prevention policy model that could guide the subsequent steps of model development.

RESULTS

Scoping Review

Our search strategy resulted in 11 indexed abstracts published between 2000 to 2020. We reviewed the abstracts and excluded 5 citations for not including a decision mode (n=4), and a language other than English (n=1). The remaining citations (n=6) were fully reviewed. Of these, two were excluded for not targeting asthma prevention as the primary outcome. The remaining 4 studies were included in the analysis. Figure 1 shows the flow of the studies in the scoping review. Table 1 summarizes the included studies. All identified models were simple decision-trees or Markov models designed to address the specific question at hand, often cost-effectiveness of a single intervention.^23–26

View this table:

• View inline
• View popup
• Download powerpoint

Table 1 Summary of scoping review of primary prevention modeling studies for asthma

• Download figure
• Open in new tab

Figure 1 Flow of analysis for the systematic review

We identified only one publication focused on the conceptualization of an asthma prevention model. Ramos and colleagues reported structuring and validating of a cost-effectiveness model of primary asthma prevention through allergen avoidance.²⁵ The authors developed the structure of the model through round-tables and validated it via further discussions with experts and comparison to other asthma models.

We did not identify any microsimulation or reference models for the primary prevention of pediatric asthma. Our conclusion from the scoping review was that none of the previous models can be adapted to act as a reference policy model for childhood asthma prevention.

Conceptualization of the Model

The initial draft of the concept map which was developed by the steering group related asthma diagnosis to biological sex, ethnicity, environment, genetics, birth mode, breastfeeding, and infections. The effects were considered to be modulated through lung function, atopy, and possibly microbiome, as shown in Figure 2A.

• Download figure
• Open in new tab

Figure 2 Initial (A) and final (B) concept maps for asthma diagnosis

The Delphi panel refined the draft concept map through two rounds of surveys and a final consensus meeting. Results of the survey are summarized in Appendix III. The panel added additional factors, renamed some factors for clarity, and clustered all factors under three domains of ‘patient characteristic’, ‘family history’, and ‘environmental factors’. The panel reviewed the direction of causality, assigned strength of relationship to each line in the concept map, as shown in Figure 2B. The final concept map also indicates that the likelihood of a diagnosis of asthma in a child is influenced by multiple factors, namely respiratory symptoms, lung function, and atopy. The panel also proposed potential measures for clinical concepts, as shown in Table 3.

View this table:

• View inline
• View popup
• Download powerpoint

Table 3 Proposed markers of the disease

It was decided that the unit of representation in the model will be individual subjects, followed from their date of birth. Given the multidimensionality of risk factors, the team recommended a discrete event simulation model. This approach was deemed particularly advantageous for building a reference model, given its ability captures complex interactions among decisions. Further, the team recommended that the model should follow an open population, due to the importance of modeling realistic aspects of rolling out an intervention or policy, such as the phased implementation of a national screening program. Given the open population nature of the model, the time horizon would be on the calendar year, instead of up to a certain patient age commonly used in previous studies. The team also decided that the model should be open-source and easily accessible to the wider research community.

DISCUSSION

Economic evaluation of asthma prevention interventions is timely, particularly because of the recent identification of early life exposures that might be causally related to asthma diagnosis later in life. Any asthma prevention strategy would require either population-level or individual-level modification of early-life risk factors that impact the disease.²⁷ Recognized risk factors range from those that might be modifiable at the patient-level (e.g. birth mode, breast-feeding, diet, pets, antibiotics, tobacco), those that might be modifiable but require change at the societal level (e.g., air pollution, socio-economic status), and those that cannot be modified (e.g. biological sex, family history, genetics). Some strategies, such as reducing unnecessary antibiotics exposure, can – and perhaps should – be adopted across the board. Other potential preventative interventions (such as modification of the microbiome) may be more suitable for high risk individuals given their cost and possible side-effects.

For clinicians, the decision problem is two-fold: who is at high risk for developing asthma and what intervention has the highest preventive potential for the individual? At the societal level and from the perspective of the health-care system, the question becomes what is the most cost-effective combination of screening and risk-reduction strategies for asthma prevention. Our ultimate aim is to develop a modeling platform for the evaluation of the value-for-money potential of such interventions and policies in a unified framework. Here, we solicited expert opinion and developed a concept map for a reference model that can assess the cost-effectiveness of any combination of risk prediction and prevention strategies.

The ISPOR-SMDM task force highlights the risk of payer’s influence on the analysis.¹⁹ However, our model is not at risk of such bias given that it is an independently funded reference model designed to address a variety of policy questions. By decoupling model development from any particular policy question, we have eliminated the risk of adopting biased assumptions that may favor the sponsor’s outcome of interest.²⁸ Further, the open-source and accessible nature of the model will facilitate independent evaluation and validation of the model.

Our systematic review of primary prevention models for asthma identified one model conceptualization²⁵ and 3 modeling publications;^23,24,26 all four were Markov or decision-tree models that modeled a single preventative intervention as a lump-sum reduction in the probability of asthma diagnosis. None modeled the underlying disease processes. In contrast, our conceptual model is based on the progression of personal and environmental risk factors, as well as atopy, lung function, and symptoms over time. Our model structure also accounts for the interplay between these factors such as modification of lifestyle and environmental risk factors in response to worsening symptoms.

This level of granularity allows for developing reference microsimulation models that can evaluate cost-effectiveness analysis of a wide range of screening and intervention methods. For example, our model can be used to inform the design features of a hypothetical asthma risk prediction score, including what target area-under-the-curve (AUC) statistic of the receiver operating characteristic (ROC) curve is good enough for the asthma risk prediction score, or whether the inclusion of additional variables, such as the metagenome and epigenetic deep phenotyping, is a cost-effective enhancement of the risk prediction model.

CONCLUSIONS

We have followed a formal procedure to solicit expert opinion and draft a concept map for an asthma prevention model. The conceptual model will form the basis for the development and validation of a microsimulation policy model for childhood asthma prevention.