Version classiqueVersion mobile
OpenEdition Books

Proceedings of the fourth Resilience Engineering Symposium

 | 
Erik Hollnagel
, 
Éric Rigaud
, 
Denis Besnard

Resilient Peer Communities

C. Alexandru, Siri Andersen, B. McKinstry, J. Ure et D. Robertson

Résumé

Increasingly “social computation” will be used in the delivery of key services, particularly public services. Social computation is a combination of computational infrastructure, voluntary and paid human effort together with an incentive structure to encourage and reinforce participation. For example, in care we will see the use of family and friends to filter out false positives from telemonitoring implementations. An important part of a social computation is the model that underpins the incentive structure. The design of such models is therefore critical if we are to enable the deployment of resilient social computations that are tolerant of new events and situations.

Note de l’auteur

Note portant sur l’auteur1

Note portant sur l’auteur2

Note portant sur l’auteur3

Note portant sur l’auteur4

Note portant sur l’auteur5

Texte intégral

1 Introduction

  • 1 “The Social Computer” takes a particular perspective on this: http://socialcomputer.eu/.
  • 2 One of Shirkey’s striking observations is that US citizens spend roughly 100 million hours watching (...)

1Over the past ten years we have seen the emergence of a new computing paradigm. Human Computation (von Ahn 2009) or Social Computation1 is emerging as a new approach to problem solving that combines computational mechanisms with human capacities (part of this capacity could come from the “cognitive surplus”2 identified by Clay Shirkey (Shirkey 2010)) via a system of incentives and reward that help shape the human response to tasks posed within the social computation. One very simple instantiation of this approach that has seen widespread adoption is “crowd-sourcing” (Doan 2011) where large tasks are split into a multitude of small sub-tasks that can be carried out by independent participants. At the moment the approach is in its infancy, however von Ahn’s ESP game (von Ahn 2009) has demonstrated that very considerable human effort can be marshalled by suitably chosen incentives and computational environment. In the longer run we believe the approach will provide the key to managing some of the large scale societal challenges that will face us over the next 20-50 years. In particular we believe that this approach will play a key role in the provision of infrastructures that many other systems use as their foundation. Careful combination of computational and human resources could make a significant contribution to the resilience of such infrastructures.

2One key societal challenge is that of ageing. In many parts of Europe as the “baby boomer” generation moves beyond retirement age we will see the ratio of employed to elderly people decline as low as one elderly person to two people in employment. In this setting, providing adequate levels of care will demand the deployment of novel approaches that are far more innovative than our current approaches to care.

3One approach that is under active consideration in the UK is the very wide scale deployment of telemonitoring to help elderly people maintain independent lifestyles well into old age. Current approaches to telemonitoring do not scale well and depend on the deployment of low-wage staff in call centres or on a massive expansion of primary care services that would be difficult to fund and staff in the current economic climate and demographic situation. Here we consider how Social Computation might be deployed as a means of tackling these issues. The goal of such a deployment would be to recruit and empower non-professionals (e.g. carers, family, friends and volunteers) to carry out many “front-line” tasks that arise in such situations. The goal of this is to arrive at a system where professionals and informal effort “co-produce” care.

4A key issue for Resilience Engineering is how to define the structure that links the computational components with the human elements of the system in a way that incentivises and rewards non-professionals for behaviour that promotes a resilient approach to changes in circumstances and avoids the creation of brittle structures that cannot adapt to change.

5The remainder of this paper considers what we can learn from a small-scale telemonitoring pilot carried out in Scotland (McKinstry 2009) and how that learning can inform the design of resilient telehealthcare systems at larger scale.

2 Social Computing

6With the advent of Web 2.0 and the rapid development of the internet towards the “future internet” and “internet of things” we have seen a huge growth of “social networking” and the development of forms of social interaction that are facilitated by the internet. This has driven a rapid growth of interest in how to develop the idea of social networking towards the notion of “social computation” where the coupling between the social and the computational is more elaborate and more directed towards achieving focussed aims. Social computing involves: some computational system that provides access to data and computational services; a social network of peers that interact with the computational system in order to carry out some task or deliver a service; and a system of incentives to encourage the behaviour of the human participants intended by the designer of the system. Incentives could be monetary, due to obligation, compulsion or enjoyment.

7A very simple example of this type of system is the winner of the DARPA Network challenge where, using financial incentives, a group at MIT created a simple system that recruited volunteers and recorded their observations in order to locate 10 weather balloons released across the continental US in around 9 hours (Greenemeier 2009).

3 Small-Scale Telecare Pilots

8As part of the preparation for a much larger telecare scheme directed towards treating a “natural-sized” care community of around 10,000 patients several small-scale pilots have been in operation in Scotland. Here we consider the experience of two of these pilots reported in McKinstry (2009) and Alexandru (2010). One pilot looked at Chronic Obstructive Pulmonary Disease (COPD) where patients can experience rapid changes in condition when acute exacerbations of their condition occur. The other pilot considered hypertension which is a much more stable condition that requires monitoring and does not have the potential for features like the exacerbations experienced by COPD patients.

9In both pilots patients were provided with monitoring equipment (e.g. a pulse oximeter that measures pulse and blood oxygen) and the means to transmit the information back to the primary care organisation. In the UK this is provided by a General Practice staffed by General Practitioners, Nursing and Administrative staff. In the primary care organisation patient information was accessed via web interfaces and was not fully integrated into the patient medical record system.

10As small scale pilots, these systems are quite resilient in the sense that the informal human peer networks surrounding the pilots can gather the necessary resources to adapt to adverse events. Here we are seeking to learn from the forms of adaptation to changes of circumstances at small scale to help identify what would be required to afford adaptation in the large scale human peer networks that will arise when telecare is deployed at national scale. Here we will focus on some specific adaptations that took place in the small-scale pilots and consider how similar adaptation could be afforded in a scaled-up system.

111. Missing readings: in the COPD pilot readings taken by patients would fail to reach the General Practice for a range of reasons. There was no systematic approach to detect failures and failure was invisible to the patient. The monitoring staff in the General Practice adapted by developing routines to check and telephone patients with missing records to ensure they had attempted to deliver the readings.

122. Continuity of dialogue: there were missing facilities to keep track of dialogue with patients and reliably recover where a dialogue had reached with a particular patient. Monitor teams managing patient dialogue in the General Practice adapted by using excel spreadsheets to keep track of dialogue and then visually link back to the main system interface to look at patient data. Additionally data was not shared with the general practice systems which meant that other caregivers could not routinely access data. This meant that observers had to either phone and tell GPs recent results or fax a print out.

133. Interpretive context: in both pilots the professional teams encountered difficulties in establishing the interpretive context for data in the system. In the case of COPD, visual cues like face colour and respiratory rate play an important role in deciding on the seriousness of an exacerbation and in the hypertension pilot difficulties arose from a lack of transparency in how some derived measures were calculated from raw data. Adaptations involved working around the system to acquire the necessary contextual information either by contacting the patient or by accessing the primary data and carrying out manual calculations. Efforts to maintain continuity of care between observers and patients also enhanced personalisation of cares and help to better contextualise data.

144. Data volume: In the COPD trails the initial design assumed that non-medical staff could carry out an initial sift of the data to identify patients who had measured over particular thresholds. They quickly discovered that this would result in very large numbers of false alarms because universal thresholds bore no strong relationship to the condition of an individual patient and most patients exceeded the thresholds fairly frequently. The pilot programme adapted by moving to setting individualised thresholds and having medical practice staff monitor for exacerbations. This led to a much closer match between the condition of the patient and the thresholds.

15Each of these examples demonstrates a successful adaptation to adverse circumstances at small scale. The qualitative analysis of the COPD pilot (McKinstry 2009) clearly indicates that patients felt they gained considerably from the pilot and that this was facilitated by significant adaptations undertaken by the participants in the pilot. Overall the aims of different groups in the pilots were quite well aligned. For example, reduction in time spent in hospital is a key factor for both medical managers, doctors and patients and there are several other aligned goals for all the involved groups.

4 Peer Networks

16A core element of a Social Computation is the peer network that structures the human contribution to the computation. This comprises a heterogeneous collection of individuals and groups with evolving needs, roles and expertise. In the case of the telecare pilots we are considering the peer network includes doctors, nurses, practice managers, call centre staff, patients, carers, family, social workers, and volunteers. The peer network uses a collection of computer-based data and services to realise a “social computation” that possibly involves quite detailed and evolving interaction between the human and mechanised parts of the system.

17The human peer network in Social Computations can draw on:

185. Technical resources: The information gathered by the system and different representations of that information; the range of actions offered by the computer-based system; and access – who can do what, where and when.

196. Social resources: informal information available in social setting; informal networks; informal interventions. Increasingly human peer networks will utilize computer-based social networking resources such as facebook, skype, email, and other forms of electronic communication to provide access to additional resources.

20Considering the adaptations we outlined in the previous section we can see that direct translation of the adaptations to larger scale contexts could be problematic:

  1. Missing readings: It is clear that there are issues around the adaptation to missing readings. It the system had been deployed at scale in this form the General Practices would not have been able to handle the task of searching for missing readings. The agenda of medical managers in the scaling up process is to support the increasing population of ageing citizens without seeing a significant increase in staffing. This suggests that different adaptations and system affordances will be required to deal with the proposed scaled-up system.

  2. Continuity of dialogue: Again the use of office tools to manage “auxiliary” information is a frequently used and often successful adaptation to design flaws and changes of circumstances for systems but as the scale of the system grows the search and linkage issues become more prominent as do slips and errors like misspellings that lead to mistaken identities and other failures of linkage between the main information system and the auxiliary data.

  3. Interpretive context: This is a widespread issue in the de-localisation of data from its context. The shift out of the local context opens up many new interpretations that would be inhibited in the local context. In small scale systems this can be easily repaired by a phone call or a walk along the corridor or a home visit but this form of adaptation is difficult to achieve in large-scale systems.

  4. Data volume: The investigators in the pilot projects identified this as an issue and made provision to “routinize” the handling of the data so that the General Practices would not be required to handle the volume of data but during the implementation it became clear that this approach could not work and the General Practices adapted the approach to one requiring much more skilled interventions. Again, this is not a scalable approach give the model where General Practices are expected to see at best a modest increase in resources to handle telecare introduction. One key observation of the telecare pilots is that the measurements taken by patients display much more variability through time and across patients than medical practitioners had expected resulting in a much higher rate of alerting.

21Experience with the pilots suggest that adaptations along these lines will be necessary in any large-scale telecare system but that the precise need for adaptation is unpredictable and so we should attempt to tackle the phenomenon exemplified in the pilot rather than “fixing” the particular issue. Our initial analysis suggests the scaled-up system should consider mechanisms that afford:

221. Marshalling of data: It is clear that throughout telecare systems there is a need to construct an orderly record from the formal data that has been gathered through the standard mechanisms and combine this with a range of other more informally gathered information that helps repair flaws and account for anomalies. It is also clear that there is a tension between the size of the task and the need for expertise and that locality or social proximity to the patient also plays an important role. There are many aspects to be considered but it seems that these three are particularly central if a scaled up system is to allow a range of adaptation in the management of data (e.g. delegating some responsibility for an individual patient’s record to the patient and their friends, family):

23a. Ordering: ensuring the record meets some standards of data quality.

24b. Annotating: the capacity to annotate records to help explain, carry context and warn that the “standard” interpretation of the data is probably wrong in this context.

25c. Provenance: people interpreting data should be able to discover how the data was generated. This information also needs to be in a comprehensible form.

262. Managing of dialogue: The whole issue of dialogue facilitated by the computer-based component of a scaled-up telecare system is an issue, in part because almost any missing channel will be replaced by some informal means (e.g. sms messages, telephone, email, ...) but also because the adaptations taken in the pilots do not scale easily. The human peer network surrounding the telecare system is quite heterogeneous and communication is at least stratified into:

27d. Communication among the immediate peer group: this is a key element in maintaining the community around the patient and to facilitate professional communication inside professional groups. It also needs to take account of informal communication and issues like locality.

28e. Communication across heterogeneous groups: This is a more sensitive issue that requires the potential to negotiate shared meaning and to engage with third parties as intermediaries. For example communication between medical professionals and patients is often mediated by friends or family.

29In designing models for human peer networks that have the capacity to adapt appropriately in a large-scale telecare system the computational components would need to incorporate elements that could deliver something equivalent to these mechanisms.

5 Models Of Care

30In a scaled-up system there are a wide range of possible human peer network structures that could support resilient delivery of telecare. One possible structure we could aim to achieve is a strong local support group of informal participants around the patient comprising carers, family, friends and the patient combined with a layered approach to the provision of professional services with clear escalation processes to trigger a professional intervention. There would also need to be the means to support multi-agency working and the voluntary sector but space constrains us to focus on just the support of the local group around the patient because this allows some illustration of the potential for incentives and learning.

31We envisage that the local group of informal participants around the patient would be responsible for maintaining an ordered record of data gathered and also potentially for alerting medical professionals in some circumstances. This group would be motivated by benevolence towards the patient (for most families at any rate) and would have access to the local context so could keep track of failures, account for anomalies in the data and would be capable of providing useful annotation. This would create a small cluster round the patient. The problem is the issue of expertise and developing the capacity to maintain the data accurately. One possibility is to have mentors from a professional group provide mentoring and training. An alternative is to use peer mentoring across groups that takes account of similarity patients condition and locality. We could further incentivise by some element of competition in the form of a data quality competition across groups that would reward groups with well maintained data. This could either be incentivised financially or by prestige. This very brief foray into social network design points to the need for more work on structures and incentives in a range of different contexts. We know from the existing social networks: Facebook, LinkedIn, etc that we can create quite different structure by varying incentives and the computational infrastructure. The question is how best to do this to enhance the resilience of mass-scale systems we depend on.

6 Conclusions

32This short paper is exploratory in the sense that it begins to consider how defining Social Computations can contribute to the resilience of large-scale infrastructures. Here we have focussed on telecare where we believe the structures will concentrate on small, local clusters to support individuals coupled to an hierarchical structure that is designed to give control over the impact on professional resources. It is clear that there are many more aspects that could be explored within the scope of Resilience Engineering. One particularly interesting area raised by a referee to this paper is the relation to Doyle’s work (Alderson 2010) on network infrastructures. From a preliminary exploration we believe this is of great interest and that the most potentially fruitful area of work might initially focus on simple crowsourced systems like the DARPA Network Challenge.

Bibliographie

References

David L. Alderson and John C. Doyle. (2010). Contrasting views of complexity and their implications for network-centric infrastructures. Trans. Sys. Man Cyber. Part A 40, 4 (July 2010), 839-852. DOI=10.1109/TSMCA.2010.2048027 http://dx.doi.org/10.1109/TSMCA.2010.204807

Alexandru, C-A. (2010). Usability Testing and Improvement of Telemedicine Websites. MSc Dissertation, University of Edinburgh, 2010.

Anhai Doan, Raghu Ramakrishnan, and Alon Y. Halevy. (2011). Crowdsourcing systems on the World-Wide Web. Commun. ACM 54, 4 (April 2011), 86-96. DOI=10.1145/1924421.1924442 http://doi.acm.org/10.1145/1924421.1924442

Paul N. Edwards, Geoffrey C. Bowker, Steven J. Jackson, and Robin Williams (eds) (2009) ‘e-infrastructures’, Special Issue of the Journal of the Association for Information Systems, (Vol. 10, Issue 5 http://aisel.aisnet.org/jais/vol10/iss5/)

Greenemeier, L. (2009) Inflated expectations: Crowdsourcing comes of age in the DARPA Network Challenge. Scientific American (Dec. 21, 2009);

http://www.scientificamerican.com/article.cfm?id=darpa-network-challenge-results

McKinstry, B. et al (2009). Reconfiguring Care for Chronic Disease with Home Monitoring: A Pilot Study Report on Telemetry-Supported Care for COPD in Lothian Region. Available at http://www.telescot.org/copd.html

Shirky, C. (2010). Cognitive Surplus, Penguin, New York von Ahn, L. (2009). Human computation. In Proceedings of the 46th Annual Design Automation Conference (DAC '09). ACM, New York, NY, USA, 418-419. DOI=10.1145/1629911.1630023 http://doi.acm.org/10.1145/1629911.1630023

Notes

1 “The Social Computer” takes a particular perspective on this: http://socialcomputer.eu/.

2 One of Shirkey’s striking observations is that US citizens spend roughly 100 million hours watching TV advertisements over a weekend of major sporting events and this is roughly the level of effort required to construct Wikipedia.

Notes de fin

1 School of Informatics

2 School of Informatics

3 Medical School, The University of Edinburgh

4 Centre for Integrated Healthcare Research, Edinburgh Napier University soa@staffmail.ed.ac.uk

5 School of Informatics

© Presses des Mines, 2011

Conditions d’utilisation : http://www.openedition.org/6540