Implementation of a computer-guided consultation in the assessment of suspected obstructive sleep apnoea syndrome

Discussion

There has recently been considerable interest in applying technologies such as CDSSs in order to improve efficiency of existing patient pathways whilst retaining the high quality of standardised guideline-delivered practice [27]. This is particularly applicable in the area of sleep medicine, where it is estimated 85% of OSAS cases in the UK remain undiagnosed [28]. Furthermore, health economic studies show that accurate diagnosis of sleep apnoea carries significant benefit, with annual NHS savings of £28 million if all the cases of moderate and severe OSA were diagnosed and managed [28]. Our data demonstrate that within a busy clinical sleep medicine service, it is possible for nonspecialists working under the clinical oversight of a sleep specialist physician to assess new referrals with suspected OSAS using a CGC quickly (each consultation lasting15 min duration or less) with the CGC maintaining a high level of diagnostic accuracy with a concordance of 93% when compared to a sleep specialist physician.

A potential concern of modifying any clinical pathway by replacing routine specialist physician assessment with nonspecialist physicians assessing patients using computer software lies in the issue of patient safety particularly where patients may present with atypical symptoms. A sleep service will commonly encounter disorders whose symptoms may masquerade as OSAS (e.g. disorders of hypoventilation, narcolepsy and central sleep apnoea) [29]. A misdiagnosis or delay in diagnosis of these conditions may result in significant patient harm. The CGC used in this study is designed to capture such atypical symptoms and alert the operator that there are unexplained features. Furthermore, the CGC requires that the operator cannot override such an alert and has to enter an action. It has been reported that if clinical staff are able to override such alerts, then outcomes suffer [30]. In our analysis, when the CGC was used by the nonspecialist staff member, there were seven cases where the CGC was not concordant with the specialist opinion but in each case, the CGC made the safe conclusion to refer for a specialist review and the software did not miss any cases of OSAS hence underpinning it's safety in routine clinical practice. However, it must also be stressed that in our clinical pathway, all consultations performed using the CGC received clinical oversight by a sleep specialist physician which also serves as an important safety net in routine practice.

The use of a CDSS in the setting of OSAS has been described previously, yet these studies have been largely questionnaire-based [31–33]. A US study of 91 subjects with a background of ischaemic heart disease (of whom 75% had OSA) found that the use of an electronic handheld CDSS carried a sensitivity of 98.5% and a specificity of 86.9%. This was compared to a commonly used screening tool, the Berlin questionnaire, which reported a sensitivity and sensitivity of 87% and 39% respectively [32]. However, none of these papers describe the integration of sleep study data into the functioning of the CDSS in terms of what was reported though this functionality had been available in one [31]. These previous studies also excluded a number of key comorbidities, such as neurological disease and opiate use, which contrasts to our system where no patients were excluded and the conclusions of the analysis of any sleep study is integrated with the history and clinical examination thus providing a more global assessment. In contrast, our study, evaluated consecutive, unselected referrals for suspected OSAS using the CGC reflecting a “real world” scenario.

Despite this, our analysis has certain limitations. Our referral population represent those patients typically referred into our service with a suspicion of OSAS and thus consisted of a cohort who had a high pre-test probability of OSAS. Such a population may carry a different pre-test probability of OSAS in comparison to a less symptomatic population referred for screening purposes (e.g. to detect sleep-disordered breathing in occupational health settings, such as in professional drivers). However, the software does screen for comorbidities which increase the likelihood of underlying OSA being present in an individual, thus prompting the user to have a lower threshold to undertake sleep studies in these situations. Furthermore, the CGC interprets the sleep study data so such patients may then be diagnosed by the software after the sleep study has been undertaken. Critics may cite the κ scores described in our analysis as showing only a modest level of agreement between the CGC and the sleep specialist. However, an explanation for this may lie in the high pre-test probability of the sample having OSA coupled with the relatively small sample size in question and clearly future studies involving the use of the software require larger numbers for validation as opposed to our current analysis aiming for proof of principle. However, from a patient safety perspective, it should also be stressed that the CGC did not recommend discharging any patient in either cohort that had unexplained symptoms in the absence of a diagnosis of OSAS but instead recommended further investigation in each case. Our study not involve performing a clinical examination in the subjects who were assessed by the nonspecialist although this features as a component of the CGC hence abnormal anatomical findings that may be indicative of underlying OSAS such as significant tonsillar enlargement and the presence of craniofacial abnormalities such as retrognathia were not accounted for. However, examination of the upper airway could be carried out by a trained physiologist or a physician following the diagnosis of OSAS and at the time of initiating CPAP and any anatomical concerns referred for specialist review. We did not perform a formal health economic impact subsequent to implementation of the CGC and such analysis is vital in any healthcare setting. However, the principle of transferring 90% of the throughput to nonspecialist physicians or primary care physicians working under the oversight of specialist physicians and doing so safely using a CGC is almost certainly a cost-effective change, although this warrants further study.

If the data here are accepted and further studies validate the safety and diagnostic accuracy of such systems, then the impact on a sleep service could be both positive and substantial. Given the significant waiting lists to access sleep services both in the UK and beyond coupled with a shortage of medical specialists, then an option to have over 90% of suspected OSAS cases assessed by nonspecialist physicians using a CGC with sleep specialists providing clinical oversight and supervision has immediate practical and economic advantages. While this happens in a measure already with specialist nurses and respiratory sleep physiologists; systems such as this may act as a robust vehicle to extend this. This could enable any healthcare system using a CGC to assess sleep and breathing disorders and correctly prescribe CPAP support or to reassure that it is not required or to indicate when onwards specialist review is needed when diagnostic uncertainty exists. Thus, implementation of technological solutions such as the CGC described here within sleep services would result in scarce sleep specialist physician time being directed to target cases where there is diagnostic uncertainty or complexity, those OSAS cases with a suboptimal response to first-line therapy and cases where the CGC had flagged up key alerts or atypical symptoms. Similarly, healthcare commissioners can be assured that the standardised assessment was comprehensive and does not generate false positives leading to the over prescribing of CPAP therapy. Furthermore, as large volumes of patients come through this intelligent system, we suggest that machine-based learning should be applied to further refine decision prompts [34]. We have described the use for initial assessment, but having collected and stored the initial data, the next opportunity is to make use of informatics for objective monitoring and follow-up of their progress, and much of that may be possible remotely.

In conclusion, we have shown that an intelligent CGC system may be used both by a specialist physician or a nonspecialist working under specialist supervision to assess patients with suspected OSAS to a point of suggesting management in line with national guideline standards. A high level of diagnostic concordance between the CGC and a sleep specialist was observed with no adverse decisions. Using “evolved” clinical decision support systems such as a CGC can facilitate assessment of people with suspected OSA allowing new management pathways to be developed.