Suffolk County Council has developed a prototype open source model which has revealed new insights about the care pathways of children in care and projects future placement needs, working in partnership with Mastodon C: https://github.com/MastodonC/witan.cic
The model and supporting data analyses address the following user stories:
- As a child in care or at risk of needing care I want the right sort of care placement at the right time to best meet my needs
- As a commissioner of placements for children in care, I want to project future demand for different types and combinations of care placement and explore ‘what if’ scenarios, so that I can achieve better outcomes for children and better value for money through strategic commissioning and service transformation.
This project will test the following hypotheses and insights and next stage developments from discovery work to date:
- Whether demand projections and simulations of ‘what if’ scenarios can lead to better meeting children’s needs and better use of resources through more informed strategic commissioning
- What time-frame of historic data best informs forward projections
- How the ‘supply’ of placements (for example different types of foster carers) can be measured as well as the ‘demand’ for placements
- Whether data from the DfE statutory return SSDA903 and other sources can be combined to provide new insights to inform service redesign and sufficiency planning
- Whether SSDA903 returns can be used to train a machine learning model to generate projections of future placement demand
- Whether other data sources e.g. case management systems can be used to further enhance demand projections
These hypotheses will be tested through further user research and model development with SCC, NCC, and CCC, to answer these questions:
- What information about the histories and characteristic pathways of children in care is important?
- What questions would commissioners and social workers like answered about expected future children in care?
- Which ‘what if?’ scenarios would Authorities like to be able to explore through modelling?
- What would be the benefits of being able to answer these questions better in achieving better outcomes for children and best use of resources?
- What sources of data are available to answer these questions? Is SSDA903 data accessible and usable with partners using the same approach we’ve used in Suffolk? What other sources of data could supplement SSDA903 data?
- What are the key features (MVP and roadmap) of a modelling solution to address these needs? How do these align with the prototype model already developed?
Local Authorities have statutory duties to protect and support children in their care. A caring home that can meet their needs is of central importance to the 99,672 looked after children in the UK. In recent years, the cost of providing such care has escalated significantly, such that this is a pressing and significant issue for every upper tier Local Authority. Provision of placements must be carefully commissioned to meet the complex and changing needs of children, to achieve value for money, and meet demand. However, children in care are highly diverse and data about them often proves inaccessible. Developing a data-driven understanding of children in care, projecting future placement demand, and exploring how that demand may vary under different conditions is therefore very challenging and not routinely or robustly undertaken.
How our hypotheses and assumptions have changed through work to date:
Work to date includes ‘discovery’ and ‘alpha’ phases. We discovered user needs through workshops with core staff including the Director of Children’s Services, social workers, the data insight lead for children’s services, data reporting staff, finance staff, and business change staff. We then developed novel data-driven insights and the prototype model with Mastodon C using agile approaches and are now refining projections and developing ‘what if’ scenarios.
The creation and analysis of a ten year joined SSDA903 dataset, previously believed to be impossible, together with new data visualisations, have revealed new insights about children in care, which can inform service improvement. We also tested whether the dataset could project future demand using a mathematical/machine learning based model.
We are now exploring different user-generated scenarios. This is revealing that core projection needs can be met with SSDA903 based information, but also that finer-grained data (e.g. varieties of foster placement) could improve the usefulness of modelling.
How we expect to change and improve journeys:
For managers seeking data-driven insights, we are making linked SSDA903 and other sources of data into usable data products for commissioners and practitioners.
For commissioners undertaking strategic commissioning, we are enabling projection of future placement demand, at a more granular level, and enabling them to explore ‘what if’ scenarios relating to activity and cost.
For social care practitioners, we expect to provide better insights to inform service redesign and decision making e.g. by visualising different types of child journey and providing detailed insights into how child characteristics (for example, age at commencement of care) can influence their journeys through care.
In 2017/18 the spend on Children in Care by Local Authorities in England was £4.5 billion. For Suffolk the figure was £42 million: nearly half the entire children’s social care budget relating to approximately 860 children. This excludes other indirect costs such as social worker time, support staff time, and court and legal costs, which are also substantial.
Children in care often have poor lifetime outcomes arising from neglect, abuse and trauma. Improving the quality of that care and investing in early intervention and prevention can improve life chances and generate wider benefits including reduced criminal justice costs, better educational outcomes leading to increased tax revenues and reduced benefit costs; and improved health outcomes, reducing NHS costs. The right care placement at the right time can help the child’s feelings of belonging and lead to earlier permanence). This is priceless.
The ability to better anticipate future placement demand using the model developed by this project is expected to lead to cashable savings by:
- Enabling more use of block purchasing of more tailored and specialist placements on a longer term basis, leading to lower costs per placement. Currently, unexpected demand leads to ‘emergency’ purchasing of placements at higher prices;
- Informing the recruitment and retention of foster care and more specialist foster care families in the right geography
- Informing the design of new interventions to reduce demand for placements, e.g. by avoiding care through targeted preventative action at key risk points identified from pathway data, and by support the sustainability of placements for more challenging behaviours.
Such service changes may be misdirected in the absence of the data-driven insights that we believe could be developed for children’s services. These benefits could be measured by comparing future costs against the model’s predicted costed case mix.
Developing an appropriate mathematical/machine learning model would cost a local authority working alone around £70k, based on investment to date from SCC and Mastodon C. There is a strong argument for developing an open reusable model that can benefit all Authorities, avoiding repeating development cost and risk, and enabling rapid shared learning.
While work to date has increased SCC’s confidence in the potential of the products of this project to enable savings and service improvements, further work is required to refine the business case. But a 1% decrease in placement costs would equate to a saving of over £400,000 in Suffolk per year, and a national saving of £40m per year.
Tools to ensure engagement and project progress:
Kick-off workshop – at the initiation of the project we will organise a 1 day workshop for staff from each LA involved in the project, to review the purpose of the project and findings from work in Suffolk to date, and to explore and expand upon user needs.
Weekly progress report – a short email highlight report will be produced each week summarising progress and next steps, to be circulated to project participants and other interested stakeholders.
Fortnightly sprint review and planning – work will be organised as a series of fortnightly ‘sprints’ (https://www.scrum.org/resources/what-is-a-sprint-in-scrum), which begin with planning of tasks for the next fortnight and conclude with a review of what has been achieved. As appropriate we will run daily ‘stand up’ meetings, via video conference, for people working intensively on tasks during a given sprint, to check progress and coordinate work for the day.
Exploration workshops – we will hold ½ – 1 day workshops including all key LA partners to explore specific questions and/or review modelling developments.
Phone/video-conferencing – for one-to-one and small group discussions we will use phone and/or video-conferencing.
Slack – we will set up a shared www.slack.com channel to support team discussions.
Github – we will use www.github.com to share code and other relevant documents.
We will establish a joint project board between SCC, PCC and CCC to oversee project progress. We would invite an MHCLG representative to join this board as appropriate.
The project board will include the Directors of Children’s Services from SCC, PCC and CCC and meet at project initiation and then once a month to completion. The project board will also be updated on progress on a weekly basis.
The project board will also include a representative from information governance and data functions in each LA to oversee data processing both by the LAs and any procured provider.
Additional support beyond grant funding will partly be provided in-house, notably from senior leaders within each of the partner Authorities. Children’s social care is not an area of public service which has historically made great use of data, and we feel there are significant opportunities to change this.
This change will require both data science expertise, but also culture change, as senior social work practitioners are not often able to draw on data to support decision making, other than basic geographical benchmarking of service performance indicators, and are therefore not used to working in a data-informed way (e.g. by forming hypotheses to be tested with data). It would therefore be interesting to explore with the Local Digital Collaboration Unit whether they are able to offer any training or skills development in the area of culture change to facilitate an environment which is open to the use and development of data-driven solutions.
We would also welcome support in facilitating informal but effective collaboration through a wiki-style community. Again, this already exists for children’s social care analysts in the Eastern region with regard to performance indicator data, but not at the level of sharing code or enabling other Authorities to use open source code and apply models in their own contexts. Support with the skills and capacity needed to enable this would be useful.