10.07.2015 Views

Systematic review, meta-analysis and economic modelling of ...

Systematic review, meta-analysis and economic modelling of ...

Systematic review, meta-analysis and economic modelling of ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Assessment <strong>of</strong> cost-effectiveness evidencereinfarction. The assumed gold st<strong>and</strong>ard for diagnosis, troponin measured 10 hours after worst symptomsis the most effective, but also the most expensive strategy because patients are admitted to hospital untilresults are available. Presentation biomarkers incur costs <strong>and</strong> may miss cases due to suboptimal sensitivity(thus worsening outcomes), but save costs by reducing length <strong>of</strong> hospital stay. We built a model to allowus to analyse the effect <strong>of</strong> different diagnostic management strategies on these costs <strong>and</strong> benefits.The decision-<strong>analysis</strong> model structureThe different diagnostic strategies were applied to a hypothetical cohort <strong>of</strong> patients attending the ED withsuspected, but not proven, ACS. We assumed that the diagnostic strategy would determine which patientshad MI <strong>and</strong> that the probability <strong>of</strong> detecting an MI was determined by the sensitivity <strong>of</strong> the diagnosticstrategy. We assumed that patients with detected MI would be managed promptly by treatment. Themodel assigned each patient a probability <strong>of</strong> reinfarction or death depending on their characteristics<strong>and</strong> whether or not they had treatment. Each patient then accrued lifetime QALYs <strong>and</strong> health-care costsaccording to their age, sex, reinfarction <strong>and</strong> treatment status. Costs were also accrued through measuringbiomarkers, hospital stay for diagnosis, further investigation, treatment <strong>and</strong>/or reinfarction depending onthe strategy <strong>and</strong> the patient characteristics. Details <strong>of</strong> each <strong>of</strong> these processes are outlined below.PopulationThe population consisted <strong>of</strong> a hypothetical cohort <strong>of</strong> patients attending the ED with suspected but notproven ACS, i.e. a history compatible with ACS but no diagnostic ECG changes (ST deviation <strong>of</strong> > 1 mmor T-wave inversion > 3 mm), <strong>and</strong> who had no major comorbidities requiring inpatient treatment (such asHF or arrhythmia). We ran the diagnostic phase model separately for patients with <strong>and</strong> without a knownhistory <strong>of</strong> CAD. Different characteristics were used for the populations with <strong>and</strong> without known CAD.Each patient entering the model had the following characteristics defined: age, sex, MI present or not,time delay between onset <strong>of</strong> worst pain <strong>and</strong> arrival at hospital, <strong>and</strong> time <strong>of</strong> day. We estimated populationcharacteristics using data from a large recent trial <strong>of</strong> point-<strong>of</strong>-care markers in patients with suspected butnot proved MI, the RATPAC (R<strong>and</strong>omised Assessment <strong>of</strong> Treatment using Panel Assay <strong>of</strong> Cardiac markers)trial. 153 Table 32 shows the population characteristics used in the model.The arrival time <strong>of</strong> patients is an important factor when considering the optimal cost-effectiveness strategybecause outside the ED medical staff may be available only at certain times <strong>of</strong> the day to make dispositiondecisions (e.g. ward rounds at specific times <strong>of</strong> the day). We analysed the arrival times <strong>of</strong> 2240 patientsfrom the RATPAC trial 153 to estimate the arrival distribution used in the model <strong>and</strong> the results are shown inTable 33. Patients in the RATPAC trial 153 presented across six hospitals over a 15-month period, so the tableis intended to demonstrate relative differences in arrival rates at different times <strong>of</strong> the day, rather thanproviding any meaningful estimate <strong>of</strong> absolute arrival rates at a particular hospital.The results are also shown in the form <strong>of</strong> a histogram in Figure 34. It can be seen that between midnight<strong>and</strong> 7 am, there are small numbers <strong>of</strong> patients. The patients arrive at a faster rate between 7 am <strong>and</strong> 9 ambut between 9 am <strong>and</strong> 2 pm is the peak time, which sees the fastest arrival rate <strong>of</strong> patients. There is a steadydecrease in the patient arrival rate between 2 pm <strong>and</strong> 6 pm <strong>and</strong> the finally, patients arrive in a constant slowstream between 6 pm <strong>and</strong> midnight.Selection <strong>of</strong> strategiesWe tested several strategies to explore the trade-<strong>of</strong>f between sensitivity <strong>and</strong> specificity. Each potentialstrategy was applied to each patient. The strategy determined:1. what tests each patient received <strong>and</strong> when2. how long each patient spent in hospital3. what treatments each patient received.82NIHR Journals Library

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!