Process Variation (Harnessing Variability, Improving Safety Part 1)

Introduction

This issue of the Human Stream examines another pillar of ‘reliability thinking’ in patient safety - namely, the question of variability.

Variability in healthcare practice, or rather the issue of ‘unwarranted’ clinical variation, has garnered increased attention in recent years due to persistent gaps between practice and the evolving evidence - mostly pertaining to the proven (or lack of) efficacy/cost effectiveness of specific diagnostic procedures, therapies and interventions.

While these are issues of care quality, unwarranted variation is routinely (and quite appropriately) framed as a patient safety issue too:

“Unwarranted variation means people are exposed to real harm from not receiving care that they need or potential harm from receiving care that they do not need and cannot benefit them.”¹

Variation in this specific sense is really about statistical variance (a measure of ‘spread’) rather than any sort of diagnosis of the underlying issue.

For instance, you could compute variance from a dataset of hospital births and determine the degree of ‘clinical variation’ with reference to an established benchmark - for example, ‘pregnancies continuing until at least 39 weeks gestation unless medically or obstetrically justified’.

While this ‘variance’ Is a form of variability, it does not explain whether variability in care practices (that is, at the process level) has anything to do with the outcomes recorded. This might seem an obvious distinction to some, but in improvement practice, we confuse the two all the time.

It doesn’t help that much of the relevant guidance on the topic of managing variation in healthcare tend to be quite imprecise around these ideas, adding to the confusion.

For instance, the Australian Commission on Safety and Quality in Healthcare, in its ‘User Guide for Reviewing Clinical Variation’² lists “Clinical care not changing in line with updated evidence” as one cause of clinical variation. If clinical variation is in essence an evidence-practice gap, then this is akin to saying evidence-practice gaps are caused by evidence-practice gaps.

Further, it is common to see guidance documents conflate outcome variation with process variation. While the ACSQH guide only does this tangentially, other resources are more direct. For example, a video presentation on managing variation, produced by the Public Health Centres for Excellence (Washington, USA)³ , makes the following assertions:

“The only acceptable variation is the one demanded by customers”.

“Quality management is about finding and maintaining the ‘one best way’. Of course, there might be many ways to do something, but there is usually one best way, and we want to standardise to that approach. Limiting variation is a key principle of quality management.”

Given the above, it is not surprising that the following logics are widely accepted (influencing everything from well-respected practice guides to many research papers):

1. That clinical outcome variation (in terms of evidence-practice gaps) and process variation (in the vein of reliability and compliance) are similar enough to be treated as one contiguous issue.

2. That variation in outcomes stem from variations in underlying processes. Put differently, that outcome reliability is attained through reduced variation in processes.

3. That variation of any kind is generally undesirable except when (and only when) variation is in response to the needs and preferences of patients.

4. That efforts at reducing such variation as synonymous with improvement.

Over the next few issues of The Human Stream we will examine, deconstruct and reframe these ideas in line with the established science and some emerging thinking.

Considering variation in two classes of systems

For sectors like manufacturing, it is wholly warranted to infer that reliable (that is consistent and unvarying) processes are necessary for producing reliable outcomes (products).

Manufacturing processes, which are fundamentally ‘technical’ (or mechanical), tend to be designed to run as highly isolated sequences of tasks with every effort made to reduce susceptibility to outside influences. Processes of this kind demonstrate linear causal relationships between process and outcomes (A causes B, B causes C and so forth).

In systems that are predominantly composed of linear processes like this, we tend to observe a high degree of correlation between variability at the process level and in outcomes. In these settings, it is safe to assume that variation in processes will proportionally manifest in outcomes and conversely, that variations in outcomes would be traceable to process variation.

However, in systems like healthcare, which have social as well as technical features, the picture can look quite different.

Healthcare is largely characterised by networked, interdependent & flexible processes - and what we regard as ‘processes’ in healthcare are rarely organised around simple causal relationships between task sequences and outcomes.

Indeed, healthcare processes are often interlaced with other processes, cooperatively sharing resources and sometimes responsibilities. As a result, these processes lack the forward and backward ‘traceability’ we expect of mechanical assemblies.

Yet, despite this seeming downside, there are several important benefits.