Big Data – A surgeon’s perspective

---

Things don’t always go to plan and sometimes complications happen. Professor Andreas Obemair explores how Big Data could make all the difference.

---

Big Data refers to the concept of very large datasets that can be linked to one or more other datasets. While Big Data is often referred to in the context of business, in this article I will focus on its potential in the healthcare environment.
Many years ago, when I was a medical student, databases were few and far between. Research involved trawling through journal articles at the university library. I physically studied countless hospital charts and extracted information from 500 patients to research the association of obesity with breast cancer prognosis. Big Data did not exist then.
And now…...
Recently, my research group showed that surgical removal of the uterus, tubes and ovaries improves survival chances in breast cancer sufferers. We identified 20,000 women diagnosed with breast cancer from the Queensland Cancer Registry and linked this data with data from the Queensland Hospital Admitted Patients Data Collection and the Australian National Death Index. Compared to 30 years ago when it took countless hours to extract data from 500 hospital charts, the time it took to obtain information from 20,000 records was minimal and the impact was much larger.
An increasing number of surgeons these days collect data on surgical outcomes. For doctors, knowledge about their complication rates has clinical benefits that have been proven across surgical specialties.
The exact mechanism by which audit improves outcomes is still poorly understood. Some experts believe that if a clinician puts a focus on health outcomes, actions consistent with achieving favourable outcomes will automatically be prioritised. Research also suggests that a surgeon’s performance may improve if they feel observed.
Data collected for clinical audit is used to measure surgical outcomes, such as readmissions to hospitals or surgical complications. By predicting clinical indicators, we can assist with the development of less troublesome procedures for certain patient groups. Data collected for audit could also be used to inform the design and even replace clinical trials as the primary source of information on the effectiveness of surgical devices in certain circumstances.
For my personal gynaecological cancer practice, auditing my surgical cases has been invaluable in improving countless clinical outcomes and patient satisfaction.
If a patient would have asked me a few years ago “Doctor, ….

• What is your overall complication rate for my procedure?


• How often do you operate on patients of my weight?


• What is your return to the operating theatre rate in patients like me?”

Questions like these would have caused me discomfort because I could not give them a direct answer. Since I now audit my surgical cases, I can say instead that “In my practice, we capture all data from surgery and its outcomes – good and bad. Specific to your situation, the overall risk of complications is x%. Because of your medical history and your body size it will be closer to y%.”
Will the average patient actually grasp the meaning of the exact numbers? Maybe not but in my experience, providing a patient with tangible complication rates is not only useful to empower the patient-doctor relationship, but is also a key ethical component of informed consent.
Limitations of Big Data
Lack of standardisation remains the biggest obstacle in making Big Data effective. For example, what actually constitutes a ‘viscus injury’ and why does a ‘serosal defect’ fall short?
In order for Big Data to achieve clear goals, those responsible for the research need to be mindful of the type of data that should be collected.
Government institutions, funders of health care (payers) and hospitals routinely collect Big Data for medical billing purposes. In the past I have been asked to review the clinical outcomes of hospital departments to advise on clinical improvement strategies. The data that I was given for analysis was the billing data described above, which some still believe is sufficient to draw clinical conclusions. Unfortunately, this is not the case, for a variety of reasons.
The data may not be broken down by procedure type.
Diagnostic laparoscopies, hysterectomies and other procedures are all mixed together and the surgical approach remains obscure, which is problematic as the expected outcomes are vastly different.
Confounders are often not accounted for. Complication rates in a young, healthy patient are vastly different to elderly patients and patients with cancer will attract higher complication rates than patients who require surgery for benign conditions.
Despite all this essential information being missing in such datasets, information about the main surgeon is available easily implying that the main driver of outcomes is the operating surgeon, which is inconsistent with current scientific knowledge.
Finally, the statistical tests that are applied sometimes lack scientific merit and normal variation of observations are often not taken into consideration.
When presented with data that is; not exact enough, leaves out crucial confounding factors, misses key sources of complications and unduly exaggerates the contribution of certain variables, how could a hospital possibly be asked to draw accurate conclusions?
Next Steps and Opportunities for Big Data
In the world of surgery there are untapped opportunities that could lead to massive health benefits, improved patient satisfaction and savings in health care expenditure.

1. 
   Standardise – Data is more useful when it is interpreted using standardised coding.


2. 
   Engagement of stakeholders – The needs of medical staff, health administrators and patients should be considered right from the beginning. All parties need to agree if Big Data is to improve health services and support research.


3. 
   Choose meaningful data – For audit, only data fields that contribute to the understanding of health outcomes should be collected.


4. 
   Apply accurate methodology – Statistical tests need to consider confounding factors and normal variation.


5. 
   Ensure confidentiality – Collection and analysis of sensitive data will only drive positive change if privacy is guaranteed for both the patient and surgeon.

Challenges to overcome
As an active surgeon with a profound interest in health outcomes, I see two main challenges for the future:

1. Will hospitals be able to engage with medical specialists and patients, seek the conversation, set expectations and respect confidentiality rather than control doctors and exert power?


2. Will my fellow surgical colleagues of various specialties be convinced that maintaining or even improving clinical outcomes is possible if we don’t even measure them? In order for clinical outcomes data to remain private and confidential without administrative control, doctors must take accountability and responsibility upon themselves.

Professor Andreas
Obermair MD
FRANZCOG CGO
Professor Andreas Obermair MD FRANZCOG CGO is a gynaecological oncologist practicing in both public and private hospitals since 2003. Andreas leads research into the predictability of surgical complications and founded SurgicalPerformance.com, web-based software for clinical audit and analytics of surgical outcomes.
ao@surgicalperformance.com

---

‘Doctor, what is your overall complication rate for my procedure?’ Questions like these would have caused me discomfort because I could not give them a direct answer.

---