Advice
The technology
The technology
ColonFlag (Medial EarlySign) is a web-based machine learning algorithm that is designed to help identify people aged 40 years or over who are at high risk of having colorectal cancer (CRC).
The algorithm uses existing datasets containing age, sex and complete blood count (CBC) test results to generate a risk score for a person. CBCs are widely used and measure the amount of red and white blood cells, haemoglobin and other factors in the blood. People with a high-risk score can be referred for further assessment, potentially before they show any symptoms.
ColonFlag works best when multiple CBCs, taken at different times, are available. The algorithm uses decision trees (similar to flow charts) and gradient boosting (which increases the prediction power of decision trees) to assess trends in the CBC parameters and patient demographics.
ColonFlag software would be installed on the central computer system within a healthcare organisation or network. The installation may be on the laboratory information system of a central laboratory or at a central electronic medical records system, such as EMIS or SystmOne. It produces a test report containing a risk score with a confirmation message.
If the score is above a defined threshold, the software will show that a person is at high risk of having CRC. The information will be passed on to the person's GP or the organisation's cancer lead. Minimal training may be needed to interpret the risk score; this is provided by the company.
ColonFlag automatically processes new data once it is saved in the healthcare network database.
Innovations
ColonFlag differs from other risk-prediction methods for colorectal cancer by automatically testing people using routinely available data. It differs from other software algorithms by using machine learning techniques to improve the accuracy of predictions.
Current NHS pathway or current care pathway
The NHS bowel cancer screening programme offers a guaiac faecal occult blood test (gFOBT) to people aged between 60 and 74 once every 2 years. In 2015, the UK National Screening Committee recommended that the bowel cancer screening programme switch to using a faecal immunochemical test (FIT) for primary screening.
For people with symptoms, NICE's diagnostics guidance on quantitative faecal immunochemical tests for colorectal cancer in primary care recommends 3 FITs. These tests are recommended for guiding referral in people without rectal bleeding and who do not meet the criteria for a suspected cancer referral outlined in NICE's guideline on suspected cancer. Other investigative tests, such as a barium enema or a flexible sigmoidoscopy, may also be done. A positive investigative test should be followed by a biopsy for diagnostic proof, and staging is done using contrast-enhanced CT.
Treatment can then begin based on the stage of the cancer.
Population, setting and intended user
ColonFlag is intended to identify, in addition to other testing methods, people who have an above-average risk of having CRC. Each software installation would cover the population of the host healthcare organisation. An administrator would be needed to operate the central computer and they would pass on the records of people with high-risk scores to either the person's GP or the organisation's cancer lead. The GP or cancer lead would then refer people to a gastroenterologist for a colonoscopy, unless contraindicated by a person's history or other risk factors.
Costs
Technology costs
Assuming that 50% of the population are aged 40 years or over (Office for National Statistics 2017), 27 million people would be eligible for testing using ColonFlag. The cost per person will decrease if the population covered per installation is higher (see table 1).
Table 1 Cost of ColonFlag
|
Description |
Year 1 cost (£; excluding VAT) |
Year 2 onwards cost (£; excluding VAT) |
|
Installation fee (1 time) |
50,000 |
0 |
|
Technical and clinical support (annual) per installation |
40,000 |
40,000 |
|
Maintenance cost* |
0 |
0 |
|
Cost per person tested** |
1 to 5 |
1 to 5 |
|
Annual cost per capita for a population of 30,000 (£5 per person tested) |
8 |
6.33 |
|
Annual cost per capita for a population of 300,000 (£1 per person tested) |
1.30 |
1.13 |
|
* Technology runs on a standard internet information services server and has no additional maintenance cost. ** The annual price per person tested is lower in larger populations. |
||
Costs of standard care
NICE's diagnostics guidance on quantitative faecal immunochemical tests to guide referral for colorectal cancer in primary care gives the cost of a gFOBT as £0.78 and the cost of a FIT to be between £1.96 and £6.04. A colonoscopy is listed as costing £372.
For people outside the screening programme's age range, the standard care would be clinical judgement in a standard GP appointment (£36, Curtis and Burns 2016).
Resource consequences
Using ColonFlag would add initial installation and implementation costs and is anticipated to increase both the number of referrals to gastroenterology and colonoscopies performed. This would need more GP resources to make referrals as well as more gastroenterologist and administrator resources to deal with more referrals. However, if more cancers were identified at an earlier stage, using ColonFlag could improve outcomes and may reduce the resources needed for cancer treatment and support. No published economic evidence on the resource consequences of adopting the technology was identified.
Minimal information technology integration is needed and basic information technology training is provided, free of charge. Basic clinical training is provided to help to interpret the risk score.
ColonFlag is not currently used in any UK organisation.