So here a revision to my prior estimate:
CEA of MDI, PUMP & AI - nominal figures, back of a fag packet
| Costs £,2020 | MDi | Pump no Sensor | AI closed loop |
| Fixed costs | Pen injector 2x£100 =200 | 3540 | 3540 |
| Recurrent Costs £ pa | |||
| Needles | 50 | na | na |
| Insulin long acting Determir £42/1500U 20U day | 204 | na | na |
| Insulin short acting Humalog | (£28/1000, 3ml vial, 20u/day) 146 | £16/1000, 10ml vial, 40U/day) 234 | £16/1000, 10ml vial, 40U/day) 234 |
| Canula | na | (3 days duration £9.80 each) 1192 | 1830 |
| Reservoir | (£2.90each, 3 days dur) 353 | (Part of 1830) | |
| CGM | na | na | 2685 |
| Test strips £15/50, 6/day | 657 | 657 | na |
| Glucagon £11.52 each | (4 pa) 46 | (2 pa) 23 | (0.25 pa) 2.88 |
| Visits £250est/ visit | (2 visits A&E pa, est) 500 | (2 *£125 paramed visits) 250 | (0.25 *£125 paramed visits pa) 31.25 |
| O’n hospitalisatn £1600/pn | (1/pa) 1600 | (0 est) | (0 est) |
| Consultancy fees £250 per visit est Diabetes &opthamology | (2*2 xpa, est) 1000 est | (2*2 xpa, est) 1000 est | (1 of each pa est) 500 est |
| One Off £ | 200 | 3540 | 3540 |
| Recurrent £ pa | 4203 | 3708 | 5293 |
| Life Years est | 45 | 50 | 55 |
| Lifetime cost | 0.1891M | 0.1889M | 0.294M |
| QALYs pa | 0.74 | 0.74 | 0.9 |
| Lifetime QALYs from diagnosis to death | 33.3 | 37 | 49.5 |
Analysis
| Lifetime | incremental Cost | incremental QALY | incremental Yrs | (Possible) Decision |
| Pump vs MDI | -£200 | 3.7 | 5 | |
| Pump dominates MDI | Pump dominates MDI | Use pump over MDI | ||
| AI vs MDI | £104000 | 16.2 | 10 | |
| Incr £/QALY ratio £6420/QALY | Incr £/LY ratio £10400/LY | Nominal values would support use of AI over MDI | ||
| AI vs Pump | £105100 | 12.5 | 5 | |
| Incr £/QALY ratio £8408/QALY | Incr £/LY ratio £21020/LY | Nominal values support use of AI rather than pump no sensor re QALY but not re clinical outcomes incremental LY ratio due to the lower nominally expected LY gain of AI over a pump(5 yr increment) vs AI over MDI(10 yr increment. Additional data may better support use of AI including for comorbidities of T1D. | ||
Table 2: Analysis Incremental Cost, QALYs and Survival
Utility information
where QALYs= utilty x time
Baseline (MDI) quality of life in UK: 0.75; https://hqlo.biomedcentral.com/articles/10.1186/s12955-015-0396-0
De Souza et al 2015. Health-related quality of life in people with type 1 Diabetes Mellitus: data from the Brazilian Type 1 Diabetes Study Group
Quality of Life Pump:0.75; various reports indicate no sig diff between pump & MDI
Eg https://academic.oup.com/jpepsy/article/31/6/650/899719, which reports HRQOL generic 0.79 and HRQOL disease spec 0.69, 0.74 was used a mean value, noting these values found in children in the US so UK adults may differ
Quality AI: 0.9; ltd info availability, AI user input used, sample of 1…
Costs were drawn from published data (BNF) where possible or otherwise estimated e.g. hospital visit fees. Quality of life was taken from publications, noting values for pump use were derived from a study on children in the US so may not be representative of UK adult values. AI qol is from a subject matter expert, an AI system user. Survival data are purely estimated based on nominal assumption. This would need checking against published data.
Interpretation
With the incremental cost effectiveness ratios (ICERs), see table, the lower values in the table, the more cost effective the item is either by producing more effect and or lower costs than to whats its compared. For example at the bottom of table, lifetime cost for pumps is expected to marginally lower than for MDI giving a better ie lower ICER ratio that is, if the value in this estimate represented real life, pumps would be more cost- and clinically effective than MDI.
But what of AI systems, such as but necessarily ltd to T1slim or Minimed 670, to MDI? The ICER for AI vs MDI is £6420/QALY, so not as cost-effective as the pump. Furthermore ai has a yet worse cost-effectiveness ratio at £8408/QALY for AI vs pump without sensor technology.
However these estimate are based on crude assumptions or nominal rather than real values, so real data may tell a diff story.
Clinical outcomes showed more variation. Estimated incremental cost/ life year were better value ie lower for the pump than MDI(ai use was predicted to be associated with lower cost, more life years) i.e. the pump dominated MDI, then AI vs MDI £10400/ extra year, than AI vs pump £21020/ extra year of life. The higher ie worse AI vs pump clinical effectiveness ratio is a product of an estimated smaller additional length of life of ai over a pump user (5 yrs) vs ai over MDI users (10 yrs). These estimates are purely nominal and real world data may vary. However user input on quality of life shows a marked improvement of ai over publ QALYs for pump no sensor or mdi. For a chronic disease, a lifetime average improved quality lends weight to the use of such a technology as ai, as indeed the incremental cost-effectiveness ratio estimate indicates.
Long term t1D morbidities’ effects and costs e.g. from haemodialysis, sight loss etc have not been included here. However as worse control would be expected with non-closed loop AI regulated management of t1d e.g. the pump no sensor or pump open loop and MDI, non-AI treatment would have higher ie worse incremental cost and clinical effectiveness ratios, supporting the use of closed loop AI systems if they indeed prove to provide high quality of life to patients and longer survival without excessive extra cost. Time will tell.
So there may be some hope perhaps depending on the figures from real studies and trials. More data will help clarify the score.
Last edited:
