Tag: predictive analytics

31 May 2017

Predictive and prescriptive modelling in health care

Following from our previous blog “How data influences decision-making in the health care industry”, we are relooking the scenario of John.

John is diagnosed with stage III colon cancer and is informed of several treatment options by his doctor. As John is not medically-trained, he does not have the medical knowledge or knowledge of new trends in treatment to decide what treatment option is the best for his specific circumstances (family history of colon cancer, 55 years of age, diabetic patient), or to decide if he needs a second opinion on treatment options. How can John make an informed decision on which treatment option to select, also considering costs of treatment which he can afford?

Further to this, was there perhaps a chance that John could have been warned ahead of time of his increased risk of developing colon cancer, given his family history, older age and the fact that he is a diabetic patient, and therefore could have been screened earlier?

Newer big data analysis techniques (predictive and prescriptive modelling) can be used to “predict” John’s risk of developing colon cancer, given his risk factors, as well as “prescribe” the most suitable treatment.

What is predictive and prescriptive modelling and how does it work?

Predictive modelling can be used to predict, for instance, the risk of developing colon cancer for a specific patient with a specific set of risk factors. This is based on past data for patients with different risk factors (predictors). Statistical regression or machine learning techniques can be used to predict a risk for a specific patient, based on the diagnosis and outcomes of other patients with similar characteristics/risk factors. The model can be updated when more data becomes available. Risk factors could also include molecular biomarkers or gene expression[1].

Prescriptive modelling is aimed at helping people make better decisions with the data at hand[2]. Prescriptive modelling uses optimisation and simulation techniques to determine all possible outcomes, as well as the best or optimal outcome[2]. Prescriptive analytics seeks to determine the best solution or outcome among various choices, given the known parameters (characteristics/risk factors). Therefore, clinical outcomes (e.g. survival due to colon cancer) can be optimised by changing inputs such as treatment provided and considering the cost of available treatments.

Who could benefit from predictive and prescriptive modelling?

As demonstrated in the example above, the patient could benefit from the insights provided by predictive and prescriptive modelling by, firstly, knowing their risk of a specific disease, and therefore getting diagnosed early, due to regular screening, if they know they are at increased risk for that disease. Secondly, the patient could benefit, by knowing what treatment will work best for him/her, also considering the cost-effectiveness of the specific treatment.

This same rationale can be used in the field of preventive medicine. Screening programmes could be aimed at those patients at most risk of acquiring a specific disease, to use the available health care budget optimally. This is also the case with vaccines. If a patient knows he/she is part of a high-risk population group, or in a high-risk phase in their life cycle, they can prevent certain diseases, like meningitis, by getting vaccinated against the disease.

Predictive and prescriptive modelling could also be used to assist pharmaceutical companies in the positioning of their products for the correct market. For instance, a specific cancer treatment may be more effective or only effective in a small subgroup of patients; for instance, for patients with a specific biomarker or gene expressed. A pharmaceutical company can use the insights provided by modelling to position their drug (especially relevant if treatment is very expensive) to only that patient group for which it is the most effective (and the most cost-effective) – a term called personalised medicine. This could help in convincing medical schemes to provide cover for that subset of patients who have the best chances of survival or cure with that drug, instead of not providing cover to any patients for that specific drug, due to the treatment being so expensive.

TCD Outcomes Research can assist you in your prescriptive and predictive modelling requirements. We can pre-process your data (if required) and then import into our modelling/machine learning platform. This data can then be analysed descriptively, predictively and prescriptively to assist with visualising the status quo, predicting new outcomes and making better decisions to benefit patients, pharmaceuticals and device companies.

References

  1. Kourou, K, et al. Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J. 2014 Nov 15;13:8-17.
  2. Fylstra, D. PASS Business Analytics Conference 2016. Available from: solver.com/files/BAMarathon_DanielFylstra_Feb25.pptx. Accessed on: 15 March 2017.

TCD Outcomes Research is a fully fledged, full service, health economics and outcomes research (HEOR) company serving healthcare companies globally and forms part of the TCD Group. We specialise in late phase health outcomes research by studying the real world value of healthcare solutions and its economic and financial impact. Partner with us to receive a skill set on a continuum of your needs, be it market access, medical, clinical, regulatory, sales or marketing. Convert scientific evidence related to efficacy, safety and quality into a market approach that focuses on real world evidence (RWE) to communicate the value of your product to your stakeholders.

23 May 2017

How data influences decision-making in the health care industry

John is diagnosed with stage III colon cancer and is informed of several treatment options by his doctor. As John is not medically-trained, he does not have the medical knowledge or knowledge of new trends in treatment to decide what treatment option is the best for his specific circumstances (family history of colon cancer, 55 years of age, diabetic patient), or to decide if he needs a second opinion on treatment options. How can data help John to make an informed decision on which treatment option to select?

Pharmaceutical company A has a new treatment for colon cancer, drug X, which they want to launch. They also want to apply for funding from the medical schemes for this drug. How can data help pharmaceutical company A to show to the medical schemes that drug X is a cost-effective option compared to the current standard of care, and that drug X should, therefore, be funded by the medical schemes? Further to this, how can data help pharmaceutical company A to know in which regions to place the bulk of their sales representatives?

Pharmaceutical company B has launched a new treatment, drug Y, and wants to broaden the indication and simultaneously broaden access to patients on a lower option medical scheme. In conjunction with this, they want to show that the drug is safe and effective in real-world practice. How can data help pharmaceutical company B to reach these objectives?

Many diverse data sets are created in the health care industry. These include patient-reported outcomes, patient registries, medical schemes claims data, clinical trial data, sales data, and more.

How can this data be used to assist John and pharmaceutical companies A and B to make informed decisions?

Firstly, the outcomes that can be extracted from health care data should be considered.

 

Outcomes that can be extracted from health care data

A diverse set of outcomes can be extracted from health care data:

  1. From patient-reported outcomes: The subjective level of joint pain experienced because of, for instance, rheumatoid arthritis, measured via a scale such as the visual analogue scale (VAS).
  2. From patient registries: The efficacy of a drug for treating a specific patient population could be determined, considering specific patient characteristics, such as age, gender, whether the patient has diabetes, as in John’s case, etc.
  3. From medical schemes claims data: Patient journeys, for instance, the in-hospital cost of treating a patient with colon cancer. Further to this, what specific areas of costs contribute most to in-hospital cost (for instance, ICU vs general ward vs medicine cost), for patients treated with drug X compared to patients treated with drug Y?
  4. From clinical trial data: The efficacy of one drug compared to another to treat a disease. This data can further be used, in conjunction with cost data and safety (adverse event) data, to ascertain whether a specific drug, such as drug X, is cost-effective, compared to another drug.
  5. From sales data: The total number of units of a drug sold per year, and factors influencing sales, such as seasonality and the location and effort of sales staff.

 

Methods used to extract outcomes from health care data

A variety of methods are used to extract outcomes from health care data:

  1. From medical schemes claims data and patient-reported outcomes: Descriptive and inferential statistical analysis using tools such as Microsoft Excel or SAS.
  2. From patient registries: Big data analysis methods using statistical modelling and machine learning, such as descriptive, predictive and prescriptive analytics.
  3. From clinical trial data: A pharmacoeconomic model can be developed that uses the efficacy data, together with cost data from amongst others medical schemes claims data, to determine whether a drug or device is cost-effective, compared to a comparable drug or device (comparator). Methods used in modelling can include Markov modelling (where a disease or treatment of a disease is broken into different states, with different utilities/weights and costs related to each state), as well as newer techniques such as discrete event simulation. Further to this, clinical trial data can be analysed using biostatistics, to prove efficacy in terms of predefined primary and secondary endpoints.
  4. From sales data: Depending on the size of the data set, different methods can be used, including big data analysis methods.

 

How can value be extracted for the patient and pharmaceutical or device company?

Methods one to three above can be used to inform John of the best drug to use for his specific circumstances, considering both the cost and efficacy of the drug and considering his age, diabetes status and family history of colon cancer (personalised medicine).

For pharmaceutical company A, methods one to four can help to provide evidence to the medical schemes of the cost-effectiveness of drug X, and to determine where to best place different sales staff members to achieve the optimal number of sales in a specific region.

For pharmaceutical company B, methods one, two and four can help to provide evidence to broaden the indication or access for lower option scheme members. These methods can also be used to prove efficacy and safety in a real-world setting, compared to a clinical trial setting.

These examples demonstrate how value can be extracted for the patient, the pharmaceutical (or device) companies as well as the medical schemes companies. However, the scientific evidence by itself is insufficient to convince them of the value of the drug. It requires a subtle combination of science, art and communication to convert these abstract concepts into value stories that will inspire them. By combining science with art, one can communicate the value of products and/or treatments in a language that appeals to each of these stakeholders. At TCD Outcomes Research, we have termed this process as “Dynamic Solutions to Dynamic QuestionsTM”.

Follow our blog for more information and case studies on data in the health care industry. Contact us to find out how TCD Outcomes Research can assist in providing you with valuable insights from your data. Visit our website for more information.

TCD Outcomes Research is a fully fledged, full service, health economics and outcomes research (HEOR) company serving healthcare companies globally and forms part of the TCD Group. We specialise in late phase health outcomes research by studying the real world value of healthcare solutions and its economic and financial impact. Partner with us to receive a skill set on a continuum of your needs, be it market access, medical, clinical, regulatory, sales or marketing. Convert scientific evidence related to efficacy, safety and quality into a market approach that focuses on real world evidence (RWE) to communicate the value of your product to your stakeholders.