The SENATOR project: developing and trialling a novel software engine to optimize medications and nonpharmacological therapy in older people with multimorbidity and polypharmacy

Introduction

Older people have the highest prevalence rates of polypharmacy, inappropriate prescribing and adverse drug reactions (ADRs) with associated high levels of morbidity [Permpongkosol, 2011; Scott and Jayathissa, 2010]. Studies from several European countries consistently show about 6% of all hospital admissions in adults can be directly attributed to ADRs [Pirmohamed, 2004; Franceschi et al. 2008; Hamilton et al. 2011]. Another large study in the Netherlands calculated that €94 million (0.5%) of the country’s total hospital budget was spent on medication-related admissions [Leendertse et al. 2011]. The cost for the UK was estimated around €706 million in 2004 [Pirmohamed et al. 2004] and €434 million in Germany in 2006 [Rottenkolber et al. 2011]. There is wide agreement that the best strategy to prevent ADR-related morbidity is to focus on high-risk groups such as older people with polypharmacy [Leendertse et al. 2011; Dequito et al. 2011]. Demand for geriatricians and other specialists in this field far outweigh supply, so the prospect of easy-to-use software to guide clinicians has tremendous potential to improve patient care. However, a major challenge is that any software solutions would need to safely handle the complexity that characterizes this patient group. A recent review concluded that there are no validated, reliable, widely used prevention strategies in older people [Petrovic et al. 2012]. Although there are a number of well-known existing tools such as the STOPP-START criteria [O’Mahony et al. 2015] and Beers criteria [American Geriatrics Society 2015 Beers Criteria Update Expert Panel, 2015], they are usually limited to use in research and are not easily applied in routine clinical practice due to the volume of information and multiple rules that apply [Caslake et al. 2013]. Some interventions provide a structured format for the review of prescribed medication, but either rely on the clinician’s considerable specialist clinical pharmacology knowledge in older people or involve applying one of the above tools [e.g. the systematic tool to reduce inappropriate prescribing (STRIP)] [Keisjers et al. 2014]. A number of Computerized Provider Order Entry (CPOE) systems are in use for electronic prescribing, but none are specially designed for older people and, though partly effective, they rely mostly on warning clinicians only about drug–drug interactions [Schiff et al. 2015]. Moreover, another potentially important strategy to avoid ADRs involves maximizing use of evidence-based nonpharmacological therapies but none of the above tools address this. Therefore, there was a clear need to develop and validate a new and more sophisticated tool that could address this important gap. Here, we describe the efforts made to date by our group to develop and test a new software engine for the optimization of medical and nondrug therapy in older people with multimorbidity and polypharmacy (SENATOR).

Establishing the SENATOR consortium

The SENATOR consortium is an international collaboration funded under the European Union FP7 programme (http://www.senator-project.eu). This followed a call for investigator-driven projects to address management of elderly individuals with multiple diseases. It is led by Professor Denis O’Mahony in University College Cork and includes collaborators from 12 European organizations with a wide range of expertise including geriatric medicine, clinical pharmacology, software design and project management.

SENATOR software

The SENATOR software incorporates a number of individually validated tools to provide clinicians with evidence-based recommendations. Key components include the START-STOPP2 criteria [O’Mahony et al. 2015] and databases of licensed indications for medications and drug–drug interactions (from the British National Formulary and a licensed CPOE product called SafeScript). The risk of death in the next year is calculated using CIRS-G (Cumulative Illness Rating Scale for Geriatrics) [Miller et al. 1992]. Another novel component is a tool that predicts the risk of an ADR, since this may influence the extent of deprescribing. Since the best current tools, such as GerontoNet [Onder et al. 2010], have only moderate ability to predict ADRs in older people [Petrovic et al. 2012], the study team plan to create, test and validate its own bespoke tool from the extensive clinical data that was being collected for the study. This will only be incorporated into the final product if it is superior to GerontoNet when validated. Provided the software is given enough information on the patient’s medical history and usual medications, it can make recommendations on inappropriate prescribing. This includes addressing under-prescribing of evidence-based treatments and recommendations for medication withdrawal to combat polypharmacy. To help keep prescribing costs down, SENATOR has information on drug availability, pricing and policies for each participating centre, and can make recommendations on the most cost-effective option. The latter involved a significant amount of original development work as this information is not easily available. The SENATOR software’s output is in the form of easy to follow bullet-point recommendations. They will still require clinicians to make their own, final judgment in conjunction with the patient themselves, as there are a number of factors that the software cannot easily take into account. In particular, there is no way of taking into account patient preference, or the result of any previous attempts at medical optimisation.

Nondrug therapies

One of the most exciting and innovative aspects of the SENATOR project is the development of individually tailored advice on appropriate non-pharmacological therapies. This aspect has required considerable original research and development as there are no existing compendia of nonpharmacological therapies, along with their evidence-based indications. This is in stark contrast to pharmacological therapies, where regularly updated tomes, such as the British National Formulary, provide clinicians with all of the information they need for safe prescribing in an easily accessible format. It is therefore perhaps unsurprising that nondrug therapies are currently under-utilized [Chen et al. 2014; Naci and Ioannidis, 2013].

The SENATOR project developed a bespoke methodology to gather the best available evidence called Optimal Non-drug Therapy for Older Persons (ONTOP; for a detailed description see Abraha et al. [2015a]). Initially, common geriatric conditions that may respond to nondrug therapy were selected for inclusion by panel discussion involving all the principal SENATOR investigators (authors). For each one, an international panel of 13 experts were asked to list and rate the clinical importance of all available outcome measures with the aim of identifying critically important outcome measures using a Delphi technique. For example, for the management of pressure ulcers, rates of complete wound healing was rated as a critically important outcome measure, whereas length of hospital stay was not. The next stage involved a ‘systematic review of systematic reviews’ of each condition, without specifying any individual interventions. This was important to avoid missing any little-known interventions. Reports were included if they assessed any nondrug therapy using systematic review methodology. From included reviews, all relevant primary studies were identified for inclusion in the final analysis. Usual inclusion criteria included: mean study population age over 65 years, randomized controlled trial design and outcomes measures that included at least one rated as critically important by the Delphi panel. This allowed the team to generate specific questions using Population, Intervention, Comparator, Outcomes (PICO) methodology to evaluate the evidence base of each intervention for each condition, and in specific patient groups. Meta-analyses were used where appropriate. Finally, results of each meta-analysis and systematic review are evaluated for quality using the GRADE methodology [Guyatt et al. 2011]. Bullet-form recommendations for inclusion in the SENATOR software are written where there is at least moderate evidence of effect. These recommendations can potentially be individualized for patients by the software (e.g. a recommendation for group exercise therapy may only be triggered where incontinence is listed as a problem, and the patient is female).

The initial SENATOR trial includes a ‘proof of concept’ study testing the feasibility of making computer-generated recommendations on nonpharmacological therapy. Initially, the trial will evaluate whether clinical teams follow advice on the prevention of delirium using nondrug techniques. The final SENATOR software will include recommendations on the nondrug treatment or prevention of 10 common geriatric conditions and some of the ONTOP reviews are already available (see Table 1) [Abraha et al. 2015b, 2016; Lozano-Montoya et al. 2016; Vélez-Díaz-Pallarés et al. 2015, 2016; Rimland et al. 2016].

Table 1.

Planned list of conditions that the SENATOR software will advise on ONTOP.

Condition
Delirium
Dementia
Falls
Heart failure
Immobility
Orthostatic hypotension
Pressure sores
Stroke
Undernutrition
Urinary incontinence

ONTOP, Optimal Non-drug Therapy for Older Persons.

Pan-European tool

One of the remits of the project was that any tool that was developed would be suitable for use throughout Europe. This represents a major challenge, as healthcare systems, therapy availability and practices vary widely. Moreover, there are significant language barriers. A major work package was devised to translate and reverse-translate all of the SENATOR user interfaces and recommendations from English to Spanish, French, Italian and Icelandic to cover the native language of all of the countries that would test the software in a clinical trial.

Testing the software

One of the major goals of SENATOR is to reduce ADR, so the rate of ADR was the obvious primary outcome for a randomized controlled trial. Although SENATOR is designed to be used in any setting, a decision was made to test it in the hospitalized setting first as this allowed more pragmatic recruitment of large numbers of volunteers at high risk of ADR within a short time period. Trials of interventions to reduce ADR pose several challenges. It is impossible to achieve effective and safe blinding of clinicians and patients. There is a risk of contamination of the control group due to clinicians learning from the intervention. Although these risks can be mitigated with a cluster-randomized design, the latter also poses many challenges. The underlying risk of ADR in different ward areas varies widely, so many different clusters would be needed to avoid an inherent bias in one arm just by chance. Moreover, out of hours cross-cover arrangements and the high workplace mobility of trainee medical staff between control and intervention units would mean the risk of contamination across clusters remained. In addition, the danger of ascertainment bias is high as ADRs in older people can easily be missed or dismissed.

In an attempt to overcome many of these inherent difficulties, the SENATOR trial was divided into two distinct trial periods. The first was an observational study across hospitals in six different countries. This allowed verification of recruitment rate and baseline ADR rate across each country and within each region in different types of wards. It also allowed the development and testing of the SENATOR software interface, a tool to predict ADR and the feasibility of data collection for all the SENATOR components ahead of the trial in Period 2. The randomized controlled trial (Period 2) will start in 2016 and will be an unblinded controlled trial with randomization at the level of the patient. Important secondary outcome measures include quality of life, length of stay in hospital, mortality and healthcare utilization costs both during the admission and at 3-month follow up.

Dissemination

The trial protocol is publicly available [ClinicalTrials.gov identifier: NCT02097654] and all major findings will be presented widely at international conferences and published in major scientific journals. Results of the main trial should become available late in 2018. If successful, SENATOR will be an extremely useful adjunct to any clinician working with older people.