National Infertility Group Report January 2013

4. What did we do?

4.1 Main Group

38. The National Infertility Group was set up, in spring 2010, at the request of Scottish Government Ministers, to help consider ways to best ensure equity and consistency of NHS infertility services across Scotland. The Group, chaired by Ian Crichton, Chief Executive of NHS National Services Scotland, met for the first time in April 2010, and thereafter met approximately every 2 months to the date of the final meeting in January 2013.

39. Membership of the Group during 2012, along with the Group's initial role and remit, is at Appendix C. The Group has made use of its members' extended clinical and patient networks as well as taking input from NHS Boards.

4.2 Working Groups

40. The Group was supported by working sub-groups which undertook more detailed research and provided expert advice on:

(1) Communication,
(2) Data Sets and Modelling,
(3) Pathways of Care,
(4) Standardisation of Access Criteria, and
(5) Single embryo transfer principles.

4.3 Review of the evidence

41. Both individual members of the Group and the five working groups sought out relevant and recent research, appraised its relevance and used the results to inform the recommendations. Where it was appropriate and readily available, the Group used Scottish and UK information sources.

4.4 Gathering information (from units, HFEA etc)

4.4.1 Data collection

42. Data are required to understand the present situation and to predict the likely future workload, costs and waiting times. The Group has carried out work on the following topics: ways of obtaining standard data; how waiting times should be defined; the development of an economic model, and the commissioning of a model to simulate the effect on numbers and waiting times of various changes in criteria for treatment.

43. Throughout the collection of data the Group has been conscious that under the confidentiality provisions of the Human Fertilisation and Embryology Act 2008¹² (HFE Act 2008), when a couple is undergoing licensed treatment, patient identifiable information from licensed centres to any person or organisation for whatever purpose, without specific consent of each patient concerned, would be a breach of the Act.

4.4.2 Obtaining routine data

44. Although Scotland has good quality national data relating to hospital inpatients and day cases, there are specific issues relating to infertility. The Human Fertilisation and Embryology legislation dictates that the only national organisation which is permitted to receive identifiable data is the Human Fertilisation and Embryology Authority¹⁶ (HFEA). Data relating to infertility, even when obvious identifiers have been removed, may still be considered a disclosure, and it is therefore not permissible for individual records to be passed to the Information Services Division (ISD) of NHS National Services Scotland¹. Although the HFEA publishes information describing the performance of the various centres in Scotland, it has no information on waiting times or costs and no plans to produce such data.

45. Data collection has been a difficult part of the process. Each of the four units collects its own data. Although the methods of data collection are not identical, they are very similar and the data group has worked to harmonise the definitions used. Based on the data needs of the economic model and the simulation model it has been possible to devise an inventory of aggregated data such that each unit has been able to contribute comparable key data. In some cases this has required a small amount of funding to assist in data collection or collation. In the future we expect better information capture within the confidentiality rules of the HFEA, and it is evident that some degree of funding will be required in future for the capture of information to assist with the 2015 review, and the Scottish Government's HEAT target. Data needs should be identified by a short life data group no later than April 2013 to enable the four centres providing NHS IVF to start capturing this information.

4.5 Modelling

46. Merck Serono Ltd¹³, a pharmaceutical company, has developed an infertility service costing model. The Infertility Network Scotland¹⁴ managed to gain ownership of this model so that it could be shared with the National Infertility Group and adapted to reflect the Scottish picture. The model is built on a decision tree process which relies on an initial feed of information on:

• Rates of arrival from GP referrals
• Proportions of cases funded by NHS, self or private funding
• Proportion of cases with good prognosis, based on age of patient
• Costs of appointments and stays in hospital.

47. Using data derived from the individual units, augmented where necessary by adapting data from England, it has been possible to create a number of scenarios and examine the cost implications over a defined period of time.

48. This approach has been helpful in examining the constituents of existing costs but less useful in predicting future costs. This reflects the limited capability of this sort of model to deal with the complexity of the overall system.

4.6 Discrete events simulation modelling

49. During the course of the work of the National Infertility Group, it became apparent that, although the economic model was important and individual NHS Boards (notably Fife) had made some useful attempts to model various scenarios using advanced spreadsheet-based work, there was a need for a deeper understanding of the relationship between changes in access criteria, waiting times and throughput. Researchers from Health Economics and Health Technology Assessment at Glasgow University¹⁵ were therefore commissioned to produce a 'discrete events simulation model' of Scottish infertility services.

50. Such a model, based on 'stocks and flows', can describe a dynamic system and allow various 'what if' scenarios to be considered and the outcomes studied. The infertility services model has been designed to undertake the following investigations:

• The evolution of the IVF/ICSI waiting list (in terms of waiting time and queue length).
• How each of the following eligibility criteria affect the waiting times associated with patients pending treatment:
  • Age
  • BMI
  • Smoking status
  • One partner has no genetic child
• How each of the following service configurations affect the waiting times associated with patients pending treatment:
  • The maximum number of allowed treatment cycles
  • The queuing method for repeat cycles

51. In particular, it was possible to use the model to work out whether it would be possible to achieve a maximum waiting time of less than one year by March 2015, and what throughput of patients would be required to achieve this target. From this, an approximate cost could be calculated on a cost per case basis.

52. The model structure was designed for each of the four NHS Level III centres in Scotland. Historical data and queue history took into account differences in criteria for each Board feeding into these centres.

53. Although this modelling is very sophisticated compared to much of the planning which is done within the NHS in Scotland, it can only ever be a relatively crude representation of reality, and like all attempts to foresee future trends, may give predictions which turn out to be inaccurate. Nevertheless, it is possible to examine some aspects of the model's strengths and weaknesses in order to gain a better understanding of the probability of its predictions coming true.

a. Structure of model. The four clinics function in very similar ways with the same basic elements (e.g. waiting times, clinics, investigations, admissions, egg harvesting, etc), and the flow of patients between the elements follows similar, and relatively straightforward, patterns. Considerable effort was made to gain sufficient understanding of the way that the system works in real life. This allowed a fairly robust model to be constructed.

b. Input data. This area caused considerable difficulty as various different types of data were needed. These are described below:

• Initial numerical data. These included such things as the number of patients likely to present for treatment and the number of patients who could be treated in a given amount of time. The data were collected directly from the units and also from the returns issued to the HFEA¹⁶ . These parameters were difficult to establish for some of the units. The model itself allowed some triangulation because the known waiting time could be used to derive plausible inputs which could then be compared to the raw data and also clarified with the data providers. We are therefore reasonably confident about these data. Where possible, 'worst case' assumptions have been made.
• Process data. It was important to try to estimate the proportion of patients who would have opted for completely private care or self-funding within an NHS facility whilst waiting times were long, but who would opt to be treated within the NHS if they knew that waiting times would be less than a year. Similarly, it was important to estimate the proportion of patients who would leave a waiting list, perhaps by achieving natural pregnancy or by moving location. In the absence of good quality data, a conservative best-guess approach had to be taken. These numbers are, however, likely to be relatively small, and so a fair amount of inaccuracy would have a relatively small effect on the outputs.
• Behavioural parameters. In particular, the smoking status and the body mass index (BMI) of patients likely to require treatment were very important because the access criteria stipulated that people had to be non-smokers and of normal BMI. Such data were not reliably known for all clinics and had to be extrapolated from one clinic to another. Any available smoking data were usually based on self-report by women, but the access criteria suggest that there should be the option of cotinine testing for both partners. This is likely to increase the number of ineligible couples compared to self-reported data.

54. It was particularly difficult to estimate the likely proportion of people who would be able to change their habits if they were smokers or were outside the BMI parameters. The data are poorly recorded and a review of available literature was unhelpful. We have tried to always err towards estimates which were likely to increase rather than decrease waiting times. Nevertheless, the BMI and smoking parameters have an important impact on the model and if the levels of these parameters are actually lower than we have estimated, or there turns out to be a greater ability to change, then there will be a noticeable increase in resource need and cost.

c. Inherent probabilistic uncertainties. The models have a probabilistic element such that the same parameters loaded into the same model and then run are likely to give slightly different results. This, of course, reflects reality; even simple structures like outpatient clinics held on different days with identical numbers of patients may take wildly differing times to complete. The models are therefore run approximately 1,000 times with exactly the same parameters and the outputs compared. The resulting waiting times form a statistical distribution from which it is possible to look at the range of variation. These ranges are presented as 95% intervals so that the most likely result is in the middle of this interval and any simulation has a one in 20 chance of being outwith the interval. For most of the simulations, in terms of waiting times, the width of this 95% range is less than 6 months. Planning has been done based on the upper limit of this range. This means that if the model projects that a one year waiting time can just be achieved, then there is only a 5% chance that this will be exceeded and it is very likely that the waiting time will be noticeably less than a year (assuming all the other model parameters and functioning is correct).

55. Where possible, the results of the Discrete Events Simulation Model were compared with the Merck Serono Economic Model. Baseline results and input parameters were broadly in line. Comparison of resulting costs between the two models for future years was difficult due to the different ways in which the models work. The Discrete Events Model gives the cost for the number of cycles that would need to be performed in order to maintain a specific waiting time and simulates individuals going through treatment under set configurations. The Merck Serono Economic model is based on a decision tree process and provides the costs for the number of cycles required if all IVF referrals were to be treated. It is unsuitable for estimating the costs of clearing the backlog of referrals but gives an estimate of costing once a steady state is achieved. This cost seems to be higher than that derived by the Discrete Events Simulation Model and this probably relates to the fact that the model does not adequately address the complexity of the system. This is borne out by the fact that the Merck Serono model suggests a considerably greater throughput once steady state is achieved. Such a throughput would suggest that there is greater unmet need at present than is observed and does not explain the relatively stable waiting times.

56. The model will be passed to Information Services Division¹ at the end of the project, allowing further simulations to be run in future using different parameters and modifications to the model. In particular, further use will be made of the model in the run up to the review in 2015 to assess timing and pace of further service improvements.

57. Further details of the outcomes of various scenarios for the Discrete Events Simulation model can be found in Appendix A, including Scenarios 2 and 8 which most closely reflect the pathway recommended by the Group.