A decision analytic model to investigate the cost-effectiveness of poisoning prevention practices in households with young children

Background Systematic reviews and a network meta-analysis show home safety education with or without the provision of safety equipment is effective in promoting poison prevention behaviours in households with children. This paper compares the cost-effectiveness of home safety interventions to promote poison prevention practices. Methods A probabilistic decision-analytic model simulates healthcare costs and benefits for a hypothetical cohort of under 5 year olds. The model compares the cost-effectiveness of home safety education, home safety inspections, provision of free or low cost safety equipment and fitting of equipment. Analyses are conducted from a UK National Health Service and Personal Social Services perspective and expressed in 2012 prices. Results Education without safety inspection, provision or fitting of equipment was the most cost-effective strategy for promoting safe storage of medicines with an incremental cost-effectiveness ratio of £2888 (95 % credible interval (CrI) £1990–£5774) per poison case avoided or £41,330 (95%CrI £20,007–£91,534) per QALY gained compared with usual care. Compared to usual care, home safety interventions were not cost-effective in promoting safe storage of other household products. Conclusion Education offers better value for money than more intensive but expensive strategies for preventing medicinal poisonings, but is only likely to be cost-effective at £30,000 per QALY gained for families in disadvantaged areas and for those with more than one child. There was considerable uncertainty in cost-effectiveness estimates due to paucity of evidence on model parameters. Policy makers should consider both costs and effectiveness of competing interventions to ensure efficient use of resources. Electronic supplementary material The online version of this article (doi:10.1186/s12889-016-3334-0) contains supplementary material, which is available to authorized users.


Background
Globally poisonings account for approximately 45,000 deaths [1] and approximately 2.4 million disability adjusted life years (DALYS) lost [2] each year in children and young people aged 0-19 years. Poisonings among 0-15 year olds have been estimated to cost the NHS more than £2 million each year [3]. In the USA, despite an estimated saving of $7-$15 for every $1 spent on poison control centres [4], non-fatal poisonings resulted in $48 million medical costs for hospitalisations and Emergency Department (ED) attendances in 2005 for the under 5 s [5].
Systematic reviews [6][7][8] and a recent network metaanalysis [9] show home safety interventions including education, home safety inspections, provision of free or low cost safety equipment and fitting of equipment are effective in promoting some poison prevention practices in families with children. However there is little published evidence on the cost-effectiveness of these interventions. This paper builds on the findings from the network meta-analysis and presents a decision analytic model to investigate the cost effectiveness of poison prevention practices in households with children aged 0-4 years.
We consider strategies that promote safe storage of medicines separately from those that promote safe storage of other household products. We customised our model for i) under 5 year olds from socio-economic disadvantaged groups 1 whom the evidence suggest are at increased risk of unintentional injury compared to those from a well-off family background [10], and ii) households with multiple children as the benefit of the intervention are likely to be greater than for families with a single child. 2 Finally, as a way of improving transparency of the modelling exercise, and prior to presenting a cost-utility analysis, we first present a cost-effectiveness analysis in which the health benefits are expressed in numbers of poison cases avoided, the natural units of the economic evaluation displayed for each intervention cohort.

Modelling framework
We developed a probabilistic decision model to compare the cost-effectiveness of seven home safety interventions to increase uptake of poison prevention behaviours in households with children aged 0-4 years. The intervention strategies evaluated in the model were identified from a recently published systematic review and network meta-analysis [9] and include: (1)Usual care (2)Education (more than usual safety education) (3)Provision of low cost/free equipment (4)Education + provision of low cost/free equipment (5)Education + provision of low cost/free equipment + home safety inspection (6)Education + provision of low cost/free equipment + fitting (i.e. free installation of equipment) (7)Education + provision of low cost/free equipment + home safety inspection + fitting We categorised the intervention strategies into these single and multi-component treatment packages based on the information reported in the individual study reports. The control intervention from individual studies was classed as usual care if the study reported the control group as 'usual safety education' , 'standard safety practice or advice' or 'no safety education' (i.e. no or do-nothing intervention control groups). Education was taken to mean that provided in addition to usual or standard safety education delivered by face-to-face contact with a trained health professional or by an educational leaflet. Free or low cost safety equipment included the provision of poison-related equipment such as cupboard or drawer locks; some interventions also provided other home safety equipment not aimed at poison prevention (e.g. smoke alarms, safety gates etc). Home safety inspection refers to home visits including inspections carried out by trained health and other professionals. Finally "fitting" refers to installation of safety equipment by a trained professional.
We constructed a cost-utility model to estimate mean costs and Quality Adjusted Life Years (QALYs) associated with the 7 strategies over a life time horizon (assumed to be 100 years). We also performed a cost-effectiveness analysis to estimate costs and consequences in the form of number of poison cases associated with each intervention over the first 5 years of life. Further technical details of the model structure and parameterisation are given in Achana [11] PhD Thesis. Separate analyses investigated the costeffectiveness of interventions to promote poison prevention practices for medicinal poisonings (ICD-10 codes X40-X44) and non-medicinal (other household products related) poisonings (ICD-10 codes X45-X49). Table 1 summarises the key features of the base case analysis. The modelled population is children aged 0-4 years, the exposure variable is safe storage of medicines or of other household products, the comparator is usual care and the outcome variable is medicallyattended unintentional actual or suspected poisoning. Safe storage is defined as storage above adult eye level or in locked cabinets and/or drawers so that it is out of reach of children [7]. The unit of analysis is the household when estimating the relative effectiveness of interventions but the individual when modelling costeffectiveness. Households were chosen as the unit of analysis in the intervention model because households are the level at which interventions are most usually provided. It is assumed that interventions act to increase the proportion of households with safe storage of poisons (at a rate determined by the relative effectiveness estimates) above and beyond the baseline prevalence of safety practices. Households with safe storage are assumed to present a lower risk of unintentional ingestion compared to households without safe storage. The aim of the decision analysis is to estimate the likelihood of an unintentional poisoning event in pre-school children for a given intervention and use this to estimate costs and consequences/benefits associated with treating such events over the lifetime of the individual. The costeffectiveness of home safety interventions will be determined with reference to the NICE cost-effectiveness thresholds of between £20,000 and £30,000 per additional QALY, where an ICER below £20,000 per QALY is generally considered to be cost-effective [12].

Model structure
A cohort simulation model is developed to estimate healthcare utilisation costs and health outcomes (number of poison cases in the cost-effectiveness analysis and QALYs in the cost-utility analysis) associated with home safety intervention compared to usual care intervention. The model consists of a decision tree and Markov model structures and is based on two previous decision analytic models that investigated the cost-effectiveness of smoke alarm give-away schemes on health outcomes in children [13,14]. Figure 1 shows the structure of the model with three distinct but interlinked sub-models: i) a first stage decision tree model referred to as the 'intervention model' , ii) a second stage Markov state transition model referred to as the 'preschool model' and iii) a third stage Markov state transition model referred to as the 'long-term model'.

Stage 1: Intervention model
In this part of the model, a decision tree is used to estimate costs and outcomes associated with the interventions being evaluated. The intervention model accounts for the baseline prevalence of safety practices in the general population, the acceptance rate of interventions in the modelled population, and the relative effectiveness of interventions in promoting uptake of poison prevention practices in the home. Costs incurred in this part of the model are costs associated with providing the intervention; a detailed description of how these are estimated is provided in the data sources section. Health outcomes in the intervention model are the number of households with and without safe storage over in the first cycle of the model. The outcomes from the intervention model serve as inputs for the second stage 'preschool' model which is described below.

Stage 2: Preschool model
The preschool model uses a Markov structure to estimate the costs and QALYs associated with each intervention strategy being evaluated in the first 5 years of life (ages 0-4 years). There are six distinct health states ( Fig. 1): safe storage (S1), no safe storage (S2), safe  storage with chronic injury (S3), no safe storage with chronic injury (S4), death from poisoning injury (S5) and death from causes unrelated to poisoning injury (S6). All individuals initially enter the preschool model in either the safe storage (S1) or the no safe storage (S2) states as determined by the intervention model. The transition probabilities that govern the movement of individuals between health states are presented in Table 2 for medicinal poisonings model and Table 3 for non-medicinal (other household products) poisonings model. These were estimated based on evidence from the literature. The two death states, S5 and S6, are absorbing states from which individuals cannot leave or move to another state in subsequent cycles of the model.

Stage 3: Long-term model
This part of the model applies to individuals aged 5 years and older. It uses a 3-state Markov structure to quantify lifetime costs and QALYs associated with chronic/long term health conditions resulting from unintentional poisonings in the 5 years ( Fig. 1). No cases of unintentional poisonings are assumed to occur during this period (i.e. at ages above 5 years; or if they do occur they are not taken into account in the model).

Implementation
The model is implemented within a Bayesian framework utilising the comprehensive decision modelling approach [15,16]. All data sources used to inform model parameters are reported in Tables 1, 2, 3, 4 and 5 and described below. Parameters are estimated by means of Markov Chain Monte Carlo (MCMC) simulations implemented in the WinBUGS software. Estimates were obtained after running three MCMC chains for 30 000 iterations using disparate starting values. The first 10,000 iterations from each chain were discarded as 'burn-in' samples to ensure that the starting values do not influence the samples on which inference is based [17]. The WinBUGS code used to implement the model is available from the authors. Uncertainty in model parameters was included by assigning probability distributions to parameters and also through deterministic sensitivity and scenario analyses. A list of model assumption is presented in Additional file 1 and the sensitivity analyses that were conducted are displayed in Additional file 2.

Data sources
Probabilities Tables 2 and 3 summarise the sources of data used to inform the model for storage of medicines and safe storage of other household products respectively. These include the baseline prevalence of safety practices in the general population, the acceptance rate of the interventions, the probability that the intervention is effective in promoting safe storage, the probability of unintentional poisonings given no safe storage, the relative risk of unintentional poisonings given safe storage, probability of using an emergency ambulance, the probability of inpatient admission, and the probability of minor, moderate, severe and fatal injury.
Costs Table 4 provides details of resource use and costs used in the model. Costs are considered from the UK National Health Service and Personal Social Services (NHS/PSS) perspective; therefore only the cost of providing the interventions and the NHS costs of treating unintentional poisoning related injuries are considered. Costs incurred prior to 2012 are converted to 2012 prices using the Bank of England inflation calculator [18].

Intervention costs
Intervention costs were estimated at the household level as a sum of the costs of resource use associated with providing the constituent components of the intervention. A detailed description of the methods and the assumptions that were made in deriving the intervention costs is presented in Table 4. For example, the cost of home safety education as a standalone intervention was estimated to be £3.67 (in 2012 prices) based on 5 min of health visitors time costing £44 per hour according to the unit costs of health and social care 2012 [19]. No travel time was added as it was assumed that contact either occurred in the clinic or that during a home visit for another reasons, for example, health visitation by a midwife health visitor. If the intervention had education and home safety inspection components, these were assumed to be provided by a health visitor during contact covering prevention of a range of injuries, of which poison prevention is a part. It assumes poison prevention education and home safety inspection each takes 5 min of a health visitor's time at £44 per hour [19]. We  Number of unintentional poisoning cases among children under 5 years in 2012 = 12029 (see Table 2 above). Forty percent of poisoning cases is due to ingestion of non-medicinal substance [33].

Utilities
The unit of health benefit (utility) in the cost-analysis is the quality adjusted life-year (QALY). Baseline utilities for non-injured individuals (i.e. those with no poisonrelated injury) were taken from general UK population utility norms [21] for people aged 18 years and above (Table 5). We assumed that individuals younger than 18 years had the same utility tariffs as the 18-25 year age range in Kind et al. [21]. Utility decrements associated with poisoning related injury were obtained from two American studies [22,23]. Miller et al. [23] reported a QALY loss of 0.03 while Miller et al. [22] reported a QALY loss of 0.046 for childhood poisoning injury (Table 5). These two figures are estimates of the utility for any poison-related injury irrespective of the severity of the injury. We assumed that minor poisoning injury were associated with the 0.03 QALY decrement (the lower of the two estimates) and moderate injury was associated with QALY decrement of 0.046 (the upper estimate). The utility decrement for severe injury was obtained by adding the upper estimate of 0.046 to a QALY decrement of 0.10 associated with the chronic injury state in the HALO study (Nicholl et al.: The Long Term Health and Healthcare Outcomes of Accidental Injury (The HALO Study) Final Report, (Unpublished)). Uncertainty around the mean utility tariffs was incorporated by assuming that the standard error of each utility decrement equals 10 % of the mean value [24]. We investigated the impact of these assumptions through sensitivity analysis.

Medicinal poisoning base case
The results of the base case cost-effectiveness analysis for safe storage of medicines are presented in Table 6. The incremental cost-effectiveness ratios (ICERs) per poison avoided increased with increasing intensity of the intervention. Compared with usual care, the intervention with the lowest ICER was education with an ICER of £2888 (95 % CrI £1990-£5774), followed by provision of safety equipment (one each of 2 types of cupboard/drawer lock) with an ICER of £4553 (95 % CrI £3284-£8892) and education plus equipment with an ICER of £6195 (95%CrI £4519-£11,030). The results of the base case cost-utility analysis for safe storage of medicines are presented in Education plus equipment and home safety inspection, education plus equipment and fitting, and education plus equipment plus home safety inspection and fitting were all associated with higher costs than more effective interventions and so were also dominated. Education was the most cost-effective intervention with probability of 0.81 followed by provision of low cost or free equipment if the cost-effectiveness threshold is £30,000 per poison case avoided. Figure 2 is a cost-effectiveness plane showing 4000 simulated ICER samples for each of the six interventions compared to usual care in the cost-effectiveness analysis (plot A) and cost-utility analysis (plot B). All ICERs lie in the north-east quadrant of the plane, suggesting that the interventions were more costly but also more effective than usual care. Figure 3 is a plot of the probability that each intervention is the most cost-effective for a range of willingness-to-pay thresholds. The plot shows that compared to usual care, home safety interventions are only cost-effective only if the willingness-to-pay threshold is about £3,000 per poison avoided (plot A) and above 50,000 per QALY gained (plot B). Table 8 displays the results of sensitivity analyses for the medicinal poisonings decision model. Sensitivity analyses for the non-medicinal poisonings model are presented in the accompanying supplementary material. Only the cost-utility results for three interventions (usual care, education and low cost/free equipment) with a non-zero probability of being the most cost-effective intervention

Non-medicinal poisoning results
Results of the base-case decision model investigating the cost-effectiveness of interventions to prevent unintentional poisonings due to ingestion of other household products are presented in Table 9 (cost-effectiveness analysis) and Table 10 (cost-utility analysis). In both analyses, home safety education interventions were associated with more costs but fewer health benefits (i.e. more poison cases in the cost-effectiveness analysis and fewer QALYs in the cost-utility analysis) than usual care. Consequently, all active interventions were dominated by usual care with the ICERs relative to usual care lying in the north-west quadrant of the cost-effectiveness plane. We replicated the sensitivity analyses listed in Additional file 1 for the safe storage of other household products model but none resulted in substantial change to the base case cost-effectiveness results.

Summary of findings
Education or provision of low cost/free equipment had lower ICERs compared to usual care than more intensive interventions for the prevention of medicinal poisonings. At a cost-effectiveness threshold of £30,000 per QALY gained, education provided to families in disadvantaged areas (£19,315 per QALY gained) or to families with more than one child aged 0-4 years (£22,960 per QALY gained) were the only cost-effective interventions. None of the interventions were found to be cost effective for the prevention of poisoning by non-medicinal substances.

Strengths and limitations
The evaluations are fully probabilistic, allowing parameter uncertainty to be taken into account in the cost-effectiveness estimates. Where uncertainty remains, for example because of model assumptions or uncertainty about which piece of evidence to use when multiple sources are available for the same parameter, these were investigated through scenario and deterministic sensitivity analyses. The results of these analyses were largely robust to many of the changes in the parameter values we tested. There are two exceptions -restricting the analysis to under 5 year olds from disadvantaged groups and increasing the number of children per household to 1.8 both increased the cost-effectiveness of home safety interventions compared with usual care. To improve transparency in our model, we presented cost-effectiveness estimated in the natural units (number of poison cases avoided) as well as in QALYs. The estimates based on the cost per QALY can help inform prioritisation of healthcare resources, whilst those from the cost per poison case avoided primarily served to validate the model predictions of the likely benefit associated with each intervention and the direction of cost-effectiveness. The main limitation stems from the lack of highquality data on the clinical history and prognosis of childhood poisoning, background utility norms and poisoning-related quality of life in children. Data on background utility norms for children were unavailable, so we extrapolated utility norms for the 18-25 year group from Kind et al [21] to earlier age groups in the model. However, preferences for health may differ between children and adults. The impact of ill health on emotional development, education and future prospects may disproportionately affect children, yet this may not be captured by utility norms obtained using instruments and valuation techniques designed for adult populations. Also, variations in utility weights across different countries have been reported for conditions such strokes   [25] and between self-and proxy-reported measures of health-related quality of life in children aged 5-14 years [26]. Hence it is possible that the utility decrements used in our model may not adequately reflect preferences for a UK injured population as they were obtained from two American studies and proxy elicited by clinicians [22,23]. Our costing assumed poison prevention education and a home safety inspection each required 5 min of health visitor time, and fitting equipment required 5 min of equipment fitter time. Safety equipment schemes usually provide advice and equipment aimed at preventing a range of injuries, not just poisonings and it is possible that we overestimated the cost of poison prevention interventions. Furthermore, our analyses do not take account of potential co-benefits from other prevention advice or equipment, for example, providing safety gates to prevent access to kitchens where hazardous substances may be stored. It is therefore possible that our study underestimates the costeffectiveness of poison prevention interventions.
Although we found that providing education was the only cost-effective intervention, it is important to remember that education alone is not the most effective poison prevention intervention [9], that there are steep social gradients in childhood poisoning [10] and that disadvantaged families often cannot afford safety equipment [27]. Reducing inequalities in childhood poisoning is an important policy objective, and our analyses did not address this issue.
Our findings are unlikely to be generalizable to countries, such as the USA, with poison control centres. These centres offer a round the clock service for the general public, providing free information and advice following a poisoning and also on poison prevention. They aim to reduce hospital visits by providing treatment at home. This is likely to result in substantial differences in healthcare resource use in countries with and without such services.
Whilst acknowledging the limitations in our adaptation of the QALY framework for evaluation of poison  prevention interventions in children, our primary motivation was to produce cost-effectiveness estimates to inform prioritisation of scarce public health resources. In the UK, this requires that cost-effectiveness is expressed in QALYs [12,28]. Even when QALYs are estimable in our analysis, they may not necessarily capture all benefits of a public health intervention. For example, parents and the economy may benefit from not having to take time off work to care for injured children, but a lack of data precluded the inclusion of this benefit in the model. Other externalities not captured by the perspective adopted in our analysis include non-health gains to other children and the emotional distress avoided of not losing a child due to poisoning for parents, grandparents and siblings. Our cost-effectiveness estimates should therefore be treated as conservative as the true benefit of the interventions may have been underestimated.

Recommendation for research
The valuation of health related quality of life in children is an active area of current research but progress is hampered by the fact that children, especially those under 5 years old, usually lack the cognitive ability and emotional maturity required to provide appropriate responses to quality of life questionnaires. A number of generic health utility instruments such Paediatric Quality of Life Inventory (PedsQL) [29,30] and child versions of the adult EuroQoL five-dimensional (EQ-5D) questionnaire [31] are available for use in studies of child health.
Currently the use of these instruments to inform economic evaluation studies of child interventions is restricted by lack of research to derive the relevant utility weights. Future research should be directed at developing strategies for eliciting the children health preferences that can be used to generate utility weights and health related quality of life information in children. In the meantime, a promising area of research in the absence of child related utility weights is the development of algorithms to derive coefficients that can be used to map the scores from an instrument where utility weights are not available to an instrument where the utility weights are available. Khan et al. [32] derived such coefficients for converting PedsQL scores to EQ-5D scores using data from a cross-sectional survey of English school children aged 11-15 years. These type of mapping exercises should be conducted on data on different clinical conditions and different age groups to assess their usefulness in health technology assessment and public health evaluations. As poison prevention interventions are often provided as part of interventions aimed at preventing a range of injuries (such as safety equipment schemes or home safety inspections), more complex decision models are needed to evaluate costs and effectiveness across a range of interventions and outcomes in a single analytic model.

Recommendations for practice
Interventions that provide equipment, plus home safety inspections and fitting of equipment are more effective than education alone. Children from disadvantaged families are at higher risk of poisoning and these families face multiple barriers, including financial barriers to making their homes safer. Reducing inequalities in child poisoning is an important policy objective, but our analyses were not able to address this. Commissioners of home safety interventions therefore need to consider our cost-effectiveness evidence as only one element in their decision making.

Conclusion
Education offers better value for money than the more intensive but expensive strategies for preventing medicinal poisonings, but is only likely to be cost-effective (at £30,000 per QALY gained) for families in disadvantaged areas or for those with more than one child. None of the interventions were cost effective for preventing nonmedicinal poisonings. There was considerable uncertainty in cost-effectiveness estimates due to paucity of evidence on model parameters. Policy makers should consider both the costs and effectiveness of competing interventions to ensure efficient use of scarce resources.
Endnotes 1 The base case analysis models representative sample of the UK households (i.e. includes a mixture of deprived and non-deprived households). 2 We assumed a single child per household in the base case analysis.