Healthcare: Optimising the treatment of heart disease

What: Singapore-based start-up Biofourmis has raised $35m in a series B round to develop its system to optimise the management of heart disease patients after their discharge from hospital. Their system, which works with a wide range of 3rd party wearable sensors, uses machine learning to determine the effectiveness of the drugs being used to treat the patients. They claim they can predict a heart failure event up to 2 weeks before it occurs with 90% accuracy. Biofourmis is still awaiting FDA approval but has run trials on 5,000 patients [1].

Commentary: With the US and other countries increasingly moving to an outcomes-based payment system for healthcare, solutions such as Biofourmis may be critical for both patients and insurance companies. However, there is no reference to the rate of false positives of their test system and if this is too high it could place a significant additional burden on healthcare costs.

[1] TechCrunch

Healthcare: A blood test for cancer

What: A start-up founded by three professors from Johns Hopkins has developed a blood test to provide early detection of five cancer types — ovarian, liver, stomach, pancreas and oesophageal – for which there are no routine screening tests for individuals with an average risk. The blood test was shown in a retrospective study to detect the five cancers with an accuracy of 69% to 98% while only having a false positive rate of 1%. The blood test now has FDA approval (for its pancreatic and ovarian cancer tests) [1].

Commentary: A routine blood test to detect cancers could lead to early detection and hence improved patient outcomes.

[1] TechCrunch

Environment: CFCs are back

What: The prevalence in the atmosphere of CFCs, the now banned chemical that was used in refrigerators and air conditioners and is responsible for creating holes in the ozone layer, has started growing again. Scientists using sensors located in South Korea and Japan were able to show that the source of 60% of the spike in CFCs was in China. Tracking the other 40% of the illegal use was more difficult due to the lack of sensors in the regions (such as Africa) where the illegal CFCs are coming from [1].

Commentary: Banning substances is one thing but policing and enforcing bans in a global world with countries that either don’t care or don’t enforce the bans is another.

[1] Axios

Telecommunications: Will 5G reduce our weather forecasting accuracy?

What: The US have started to auction 24GHz spectrum for use by 5G radio networks. Despite having five years notice of the impending auctions, at the last moment the weather bureau and NASA have objected saying that it “would result in about 77% data loss from passive microwave sounders”, the devices which operate at 23.8GHz and are used to sense the insides of weather systems in order to improve the accuracy of weather forecasts. The weather bureau claim that their forecasting accuracy will be dragged back to the 1980s by the use of the 24GHz band for telecommunications [1].

Commentary: 24GHz is almost in the millimetre wave range and the usefulness of these frequencies for outdoor use in 5G is still unknown (it could be an effective frequency for in-room applications). It is thus hard to understand at this point the validity of the weather bureau claims.

[1] Axios

Healthcare: Electronic medical records are not a panacea

What: Clinical Decision Support (CDS) Systems, often based on AI, should be able to assist doctors in avoiding medical errors, determining the optimum treatment for a specific patient’s condition, and identifying the critical pathology tests that should be conducted. However, they are dependent on the data supplied to them via the electronic health record system. Research has shown that this data has a ‘half-life’ or a mere 4 months due to rapidly changing treatment regimes and is often wildly inaccurate due to poor record keeping by physicians. As a result, doctors either ignore or override the CDS systems making them of little value [1].

Commentary: Poor data quality was blamed for the highly publicised failure of IBM’s Watson healthcare system. The medical field is not renowned for its meticulous record keeping (after all, most doctors seem to have illegible handwriting) which may make implementing machine learning in this environment very challenging.

[1] Axios

Retail: Optimising sales using the weather

What: The weather can affect retail sales in obvious ways such as fans sell more rapidly in a hot weather spell. However, by combining retail sales data with weather data, IBM have shown that there are much more subtle influences that the weather can have on retail sales. For example, pop tarts sell well before hurricanes! IBM are selling their analyses to retailers in order to enable them to optimise inventory using weather forecasts [1].

Commentary: Machine learning based on the right data (which includes not only the right parameters but also the data accuracy) sometimes allows subtle patterns to be uncovered, which is where the myth that machine learning can be predictive comes from. If the patterns are changing over time, as in the medical records example above, the so-called predictive power diminishes with the timescale of the change in the data.

[1] Axios

Cybersecurity: How much is your personal information worth?

What: Your stolen/lost personal information can sell for surprisingly little on the dark web. According to Experian, a credit card number can sell for as little as $5, and even with the CVV number it is only worth about $15. Even online payment services login information (e.g. Paypal) can sell from $20 to $200. US Passport details on the other hand are worth up to $2,000. The value of the information is determined by supply and demand as well as how many times the information can be reused before it is shut down [1].

Commentary: Of course, this information comes from Experian who sell identity protection products so should be taken with a grain of salt. Also, while a stolen credit card number may not be worth much to thieves it is worth a great deal to the consumer in terms of time and frustration to recover from the theft.

[1] Experian

Professor Hugh Bradlow is President of the Australian Academy of Technology and Engineering. He is recognised as a global leader in telecommunications technology, including being named by Global Telecom’s Business in the top 100 most powerful executives in the global telecoms Industry two years in a row, and by Smart Company as one of the 12 most influential people in Australian ICT.