How we use data analytics to maximise patient safety

Patient safety has always been a priority for clinicians, policymakers, and regulatory authorities. And for good reason, as without adequate enforcement of safety regulations, other interests could jeopardise the essence of healthcare. 

Despite the general upward trend in healthcare safety standards, recent media coverage would suggest these standards are under threat of significant regression, particularly when it comes to telehealth.

A key concern is that asynchronous services allow a large number of people to obtain prescription medications that are a risk to their health and well-being — but is this true?

Read as we explain how asynchronous care models can actually improve patient safety across the health system, and how we do so with Juniper’s Weight Reset Program. 

What we know about telehealth safety

Digital healthcare isn’t new. 

It’s been around for more than a century and developed steadily over the years with the advent of various technologies [1]. However, nothing came within even a fraction of the surge experienced during the COVID-19 pandemic.

The outbreak of the virus meant that turning to digital healthcare was the only way of maintaining regular health services while complying with social distancing rules and managing hospital resources — a new norm that exacted an abundance of research and studies on telehealth safety.

Now, some of this research has its limitations. For example, most patient safety reviews only look at direct markers of safety, excluding indirect markers like travel, even though telehealth’s digital nature helps overcome geographical barriers and save more lives on the road [2]. Plus, the bulk of literature on telehealth safety appears to limit its scope to synchronous care.

Nevertheless, enough evidence has been generated to draw several conclusions:

• Treatment adherence, symptom surveillance, and appointment attendance improve when patients receive text-based care [3] [4] [5] [6] [7].
  
• In certain circumstances, chatbots are capable of more empathetic and better-quality responses to patient questions than humans and can safely perform patient monitoring tasks [8] [9] [10] [11].
  
• Patients and clinicians typically report a high level of satisfaction with asynchronous treatments [12] [13] [14].
  

However, the findings that arguably best capture the safety potential of asynchronous care are those from the 2023 OECD publication on Integrating Care to Prevent and Manage Chronic Diseases. 

With the goal of identifying the best systemic methods for reducing care fragmentation and reversing chronic disease trends, the report highlights 6 key finding dimensions, 4 of which are relevant to asynchronous care safety: digital tools and health information systems, monitoring and evaluation, health equality, and scaling-up and transferrals.

The report goes into detail about the role that quality data plays in each of these 4 dimensions — a role that becomes quite clear when we consider their direct goals. For example, better data leads to more accurate monitoring, more equitable access for disadvantaged groups, and more efficacious up-scaling initiatives.

How Juniper’s healthcare model works

Since day one, we’ve been applying learnings from the above-mentioned report in our healthcare model. 

A few ways we do so, particularly in the prescription phase (the prevailing theme of recent media criticism), include:

Our pre-consultation questionnaire

To start their Juniper journey, every patient completes a detailed pre-consultation questionnaire — so detailed that it can have over 100 questions, depending on the person’s responses and their specific condition.

Primarily, these questions are designed to elicit a comprehensive medical history for each patient. 

But they’re also a starting point for targeted consultations with our clinicians, who can then request additional information via phone, video, or our platform — such as photos, blood tests or external validation of a patient’s medical history — to ensure safe clinical decisions are made.

As a result of this thorough filtering process, almost 40% of patients who enter a consult are deemed unsuitable for our services. Instead, they’re referred to a local GP, who can be provided with all the new information collected in the questionnaire.

Our ACHS certification

Our doctors are bound by the same rules and regulations as every other Australian GP, and for patients who are recommended medication as part of Juniper’s Weight Reset Program, scripts are only ever issued by our Australian-accredited doctors.

Our certification from the Australian Council on Healthcare Standards (ACHS) is a reflection of our commitment to national regulation — an accreditation that, to our knowledge, no other telehealth company in Australia can boast.

Our auditing processes

Like community GPs, our doctors are human, and therefore prone to human error. However, unlike most community GPs, they’re supported by our central data repository’s algorithms, which intercept errors before they impact the patient.

The level of risk in misprescription exists on a spectrum and response matrices will vary depending on the scale of urgency:

High-risk decisions:

• Are more likely to cause immediate harm
• Need immediate intervention, which relies on timely detection

Low-risk decisions:

• Are vulnerable to oversight due to their indirect effect on adverse events
• Are often harmless, but predictive of future adverse outcomes. If certain prescribers start to normalise decisions with a level of risk that is low but significantly higher than the cohort average, they likely become more susceptible to high-risk decisions in the future
• Require a rigorous system for detecting any patterns that can predict more serious risk

Our central data repository enables us to address all ill-advised prescription decisions across the risk spectrum through 2 auditing processes. 

1. Reactive: To target prescription decisions with a ‘high-risk’ rating

Every 24-72 hours (depending on potential patient harm), our data repository produces review datasets from the 18 queries that our auditing team created to identify high-risk prescribing.

This means, for example, that the Euc auditing team would be alerted in cases where a patient with a history of pancreatitis was prescribed a GLP-1 RA.

From there, a robust incident response process is followed. Once the issue is intercepted, the team conducts a thorough safety and quality review of the implicated prescriber to determine management and escalation.

As of 23 June, our doctors have prescribed medication to 157,105 patients in 2023, of which only 86 have been identified as errors. This converts to a misprescription rate of 0.05%, a number that outperforms public health system outcomes, but we’ll get to that shortly.

2. Preventative: To identify ‘abnormal prescriber patterns’

The second process exists to prevent future high-risk decisions by detecting outliers in markers of prescriber behaviour.

For example, our dashboards highlight clinician prescription rates that are significantly higher or lower than their peers and therefore may indicate lower compliance with our clinical guidelines. Plus, weekly consults are monitored to prevent doctors from exceeding workload standards and compromising decision-making quality.

Although abnormal prescriber patterns may not have contributed to any misprescription, they can reflect a lower standard of attentiveness that may lead to higher-risk decisions in the future. Auditing prescriber patterns also allows managers to provide more objective feedback, which facilitates continual professional development. 

How does this compare to safety standards in face-to-face settings?

The best way to answer this question would be to compare our high-risk decision rates with public system misprescription figures. 

The most reliable report on this subject is a 2022 review on The Extent of Medication-Related Hospital Admissions in Australia, which triangulated data from 17 studies between 1988 and 2021 and found that “between 2 and 4% of all (hospital) admissions were medication related, with rising estimates when sub-populations were studied.” [15]

Although these “medication-related problems” included “under- or overdosing” – a factor largely outside the control of a prescribing doctor – this typically accounted for less than ¼ of problems, with “adverse medicine reactions” being the most common type of admission.

Plus, the authors stressed that their data-driven estimates were conservative and didn’t consider “medication-related emergency department presentations that did not lead to hospital admissions,” which they believe would have added 400,000 annual presentations to their final estimate of 250,000 (of which ⅔ were potentially preventable).

But even applying the most conservative estimate of ⅔ of 2%, there’s still a misprescription rate of 1.33% – a rate over 24 times as high as the one observed among Eucalyptus doctors in 2023. Moving towards the 4% figure and adding a portion of the 400,000 less severe emergency presentations would see that rate explode.

This disparity may not come as a surprise when we consider the standard of care continuity in face-to-face GP clinics in Australia. 

We explored this topic in detail in a recent article, but long story short:

• The RACGP has claimed that regular GPs “ensure continuity of care” by having complete access to patient medical history, yet current literature suggests otherwise [16].
• In 2012, the Australian government rolled out My Health Record, a database where patient information could be centralised, but its implementation has been undeniably poor, which has greatly hindered care coordination and continuity – particularly for patients with chronic conditions.
• Our national health system has been labelled “considerably poorer in patient engagement and delivering preventative, safe and coordinated care” than other OECD nations [17].

Considering all of this, it would be a no-brainer for the Australian healthcare system to try and emulate the core feature of our care model: our data repository.

Despite recent criticism, our care model is our greatest strength in care continuity and patient safety for one simple reason: it automatically uploads all patient data to a central repository, enabling our analytical tools to pre-empt and detect our doctors’ errors. 

We’re not suggesting that all healthcare be delivered asynchronously. We’re simply arguing that based on our findings, care quality and safety outcomes would improve immensely if our national health system developed an efficient means for uploading and centralising patient data similar to what we have at Juniper.