Community-acquired pneumonia (CAP) is one of the most common diseases among older people, and is a leading cause of hospitalisation and death in the elderly population.1^–3 A number of studies have identified specific interventions that can improve outcomes in the treatment of patients hospitalised with pneumonia. These studies have shown that timely initiation of appropriate antibiotics and vaccinations is associated with improved clinical outcomes, such as shorter hospitalisations and reduced mortality.4^–9 Unfortunately, not all patients who are hospitalised for CAP receive the right care at the right time.71011

The US Centers for Medicare and Medicaid Services (CMS) has developed performance measures that address antibiotic timing, diagnostic testing, appropriate antibiotic selection, vaccinations and smoking cessation counselling for eligible hospitalised patients (see table 2 below).11 A number of improvement efforts related to CAP care have been published in the literature, and varying levels of success have been reported. These studies describe interventions to modify care, such as teleconferences, regional seminars, physician pocket cards, newsletters, clinical case management, standing orders for vaccinations, clinical guideline posters and standard admission orders.12^–20 Although some of the articles reported improvements in the quality measures, other articles did not show substantial improvements. Attempts to introduce improvements into healthcare systems, which can include standardisation of care and reduction of unnecessary variation, can be met with scepticism and failure of adoption by providers.21^–23 Although there are many reasons why interventions may not be embraced by providers, one concern is that increased standardisation of care reduces provider and patient autonomy and judgement that is needed to address complex aspects of patient care. Of the improvements that were successful, it is not always clear which interventions were considered and which best address each of the clinical measures and/or the local context in a particular setting.

Glouberman and Zimmerman have described a typology of simple, complicated and complex problems.24 A simple problem, such as how to bake a cake from a box, is best addressed by following a recipe or instructions. A complicated problem, such as how to send a rocket to the moon, can contain subsets of simple problems but they are complicated because of the scale of the problem and the need for coordination of resources or specialised expertise. A complex problem, such as how best to raise a child, can contain simple and complicated elements but it has the unique characteristics of interdependency, non-linearity and the ability to adapt as conditions change.2425Table 1 summarises the three problems and selected characteristics of each.

We believe that clinical care for individuals and populations also involves simple, complicated and complex problems, and clinical realities. This article describes the use of this typology to assess and understand the results of interventions to improve the care of patients with pneumonia. This framework has helped us assess and reflect on the “fit” of the interventions with the particular reality that was encountered.

LOCAL PROBLEM

Dartmouth-Hitchcock Medical Center (DHMC) is an academic medical centre in Lebanon, New Hampshire, USA. The hospital has a 396-inpatient bed capacity and serves as a major tertiary care referral site for northern New England.

In response to measured unreliable care (with respect to CMS performance measures) for hospitalised pneumonia patients, a multidisciplinary CAP Improvement Group was formed in December 2003. Table 2 displays CAP quality measures for DHMC compared with other academic institutions (University Healthcare Consortium); these measures showed variable results and the composite (or “all or none”) measure was 10% at the initiation of the improvement work. Performance measures in the 30–60% range reflected care that was heavily dependent on individual provider discretion, knowledge, actions and/or clinical judgement.

INTENDED IMPROVEMENT/AIMS

The CAP Improvement Group developed the global aim to improve the quality of care for all adult patients diagnosed with CAP and hospitalised at DHMC, with the specific aims to (1) get correct treatment reliably started in less than 4 h after arrival and (2) prevent unnecessary pneumonia by improving immunisation delivery to and smoking cessation for eligible patients.

METHODS

Prior to initiation of any interventions to improve CAP care, the improvement team assessed the process of care. This assessment revealed that one of the main barriers to timely administration of antibiotics was the “handover” in patient care during the transition from the outpatient setting (the outpatient clinics or emergency department) to the inpatient care unit. The “handover” occurred at a critical point in the patient’s illness, just after diagnosis, but before initiation of treatment. This process of care involved three different types of clinical microsystems: the outpatient clinics, the emergency department and the inpatient units. Microsystems are defined as the small group of people and information who regularly work together to provide care to discrete subpopulations of patients. They are the functional settings where patients, families and healthcare meet.26 The variable timing and nature of the “handover” in patient care between these traditionally independent microsystems often led to considerable delays and variation in the treatments given in the initial care for CAP patients. Furthermore, as patients could present through several different clinics (along with the emergency department) and could be hospitalised in several different units throughout the hospital, a large number of potential interactions and distinct processes of care were involved in any given admission process and handover in care.

Prior to the implementation of improvement efforts, the outpatient care units worked independently and believed that their responsibility for a patient ended when a patient was admitted to the hospital, even though they may have physically remained in the outpatient care unit. The outpatient providers saw their role as making a diagnosis and determining a destination for the patient. If a patient needed hospitalisation, the initial treatment plan was typically deferred to the inpatient providers, even if the patient stayed physically present in the emergency department or clinic awaiting an inpatient bed. The process of admission, awaiting an open bed, transfer to an inpatient unit and re-evaluation by the often-busy admitting team of resident doctors all contributed to unnecessary delays in treatment. As a result of this pattern of care, treatment was often delayed for ⩾8 h after a diagnosis of CAP was made. Reframing these different care settings as a common system of care for patients seemed a clear first step towards improving the care for CAP patients.

In assessing the other quality measures for CAP, it was discovered that there was no systematic process to ensure that patients received appropriate diagnostic tests or preventive care. The process was dependent on individual providers to ensure that patients received appropriate diagnostic testing and to remember to order vaccinations or smoking cessation referrals for eligible patients prior to discharge. As a result, these measures varied widely depending on the particular conditions or providers who cared for the patients, resulting in unreliable care.

PLANNING AND IMPLEMENTING THE IMPROVEMENT WORK

To address the variation in initial antibiotic selection and the considerable delays in initiating antibiotic treatment, two separate order forms were created, one for use in the emergency department and outpatient clinics and the other for admission to the inpatient units. These order forms listed diagnostic tests that should be performed prior to treatment, and prompted antibiotic ordering from a list of recommended antibiotics, specifying the timing of the initial dose. Documentation of care received on the order form was designed to improve the communication between the outpatient and inpatient settings. The back of the emergency department order set included an algorithm based on patient factors and clinical status to assist in the identification of patients with a low mortality risk who might be best treated without hospitalisation.

In addressing vaccination administration and smoking cessation, the improvement team decided that a checklist of orders embedded into the standard admission forms for doctors and nurses would be the best way to introduce a “forcing function” to ensure reliable care. In addition, a standing order was developed at the institution to allow all nurses to order and administer vaccinations for eligible patients without a specific doctor’s order. Process changes for doctors and nurses using the standard admission forms and standing orders for vaccinations were made to the admission and discharge process so that all hospitalised patients (not just CAP patients) received recommended preventive care prior to discharge.

The third intervention was the development of a real-time feedback system to providers. Each time a CAP patient was admitted, the medical record was reviewed within 1–2 days to determine whether the CAP performance measures had been met. A summary of the measures was sent by email to the providers involved with the patient’s admission. This feedback initiated a dialogue between the improvement group and the treating providers that encouraged identification of system issues that prevented delivery of reliable care and also identified complex patient issues in which an ongoing flexible, informed provider–patient relationship took priority over meeting performance measures.

METHODS OF EVALUATION

The CAP data were collected by data abstractors who collected patient data on a quarterly basis to submit to UHC and CMS and for the institution’s public website, which included performance measures. Measures were abstracted and plotted monthly in statistical process control charts. Individual medical records of patients were reviewed on a regular basis by the CAP Improvement Group. In addition to the previously identified quality measures, the percentage of patients who were admitted using the standard admission orders and qualitative feedback from providers triggered by the email feedback were reviewed on a regular basis.

ANALYSIS

To determine whether there was substantial progress across 4 years (2004–7), a χ² statistic in a 4 × 2 contingency table was used. Although the quality measures were tracked in statistical process control charts, a traditional statistic was used for concise summarisation for this article. Year 2004 was considered the preintervention year while year 2005 was the first year of the implementation of the improvement interventions. In year 2006, further adaptations to the interventions and process changes addressing vaccinations and smoking cessation for hospitalised patients were implemented, and year 2007 can be considered whether results from the interventions were sustained over time. χ² was also used to assess whether proportions in 2005, 2006 and 2007 were different from 2004. A p value <0.05 was considered significant.

RESULTS

Table 3 displays five of the six CAP measures (listed in table 2) over the 4 years and the χ² results. Oxygenation assessment was not analysed since it was 100% over the 4 years. Statistically significant improvements were achieved in four measures.

DISCUSSION

As we reflected on the implementation of the interventions and the associated changes in measures, we realised a connection between our thinking and an approach to healthcare system problem analysis first described by Glouberman and Zimmerman.24 Simple scenarios, such as measuring vital signs and administering vaccinations, usually have a few well-known contributing factors that form the treatment plan and determine relevant outcomes. The clinical uncertainty is typically low and the outcome of their contribution to the overall care is usually known or predictable based on available scientific evidence. Patients should understand the treatment plan and associated risks and benefits but more active elicitation of patient preferences is not likely to add to the quality of the care. As these care components involve little scientific disagreement or benefit from patient preference determination, they are associated with relatively low levels of patient and provider autonomy and judgement. Improvement interventions most suited for these situations are typically forcing functions, checklists and standard orders. These situations usually involve “all or never” or “yes or no” clinical decision points. As seen in this improvement effort—measures in the upper 80% range were achieved with the use of a standard admission form that included a checklist of preventive care, a standing order for vaccinations and process changes to the admission and discharge process. In addition, specific doctor involvement in vaccination and smoking assessment was eliminated or greatly reduced with the new process changes. This effectively reduced individual provider autonomy and discretion with these treatment decisions while increasing the reliability of the care.

Complicated scenarios, such as determining the ideal antibiotic regimen for a patient or determining the ideal treatment location, have more background information and factors that help determine the treatment plan and outcome. These factors are usually discoverable through detailed knowledge of a patient and/or a clinical situation. The clinical uncertainty about the treatment plan and expected outcome can vary based on the situation but is generally knowable and largely predictable based on the fit of the scientific evidence and guidelines to the specific patient scenario. Patient involvement becomes increasingly important in this realm, as patients must understand the facts, risks and benefits of planned treatments, and their values become integrated into the treatment plan. These situations are ideally suited for a clinical pathway or treatment algorithms (“if/then” decision trees) that take into account the contingent nature of the situation manifest in the specific patient characteristics, the scientific evidence and the treatment setting.

Complex scenarios have multiple and possibly conflicting background factors and comorbidities that make the development of a treatment plan more difficult. For instance, consider a 65-year-old patient who has a history of heart disease and lung cancer and is admitted with an acute myocardial infarction and pneumonia. The patient becomes increasingly ill during the admission process and his vital signs and oxygenation status deteriorate. His medical record also contains a “do not resuscitate” order and his wife confirms that the patient reported to her that he would never want to have “life support”. Available scientific evidence would suggest that treatment in the intensive care unit and prompt treatment of the pneumonia and the myocardial infarction would represent best practices in this case. However, such evidence is derived from studies that may have included a very different patient population and must be carefully scrutinised for applicability to this particular patient’s medical history and his stated wishes and values. In addition, standard guidelines and recommendations must be adapted to the individual patient’s situation and treatment goals. A clear understanding of a patient’s wishes, goals and values may be the most important factor in the development of a treatment plan. When elements of the care situation can only be partly known and where their inter-relations can change, the outcome is largely not predictable. In these situations providers need to work to create a shared aim and to work through good relationships. This is usually associated with the need for high degrees of both provider and patient/family autonomy, flexibility and respectful interaction.

While the ideal improvement aim with regard to the safety of simple and complicated situations was to improve reliability, the improvement aim for complex situations was improved resilience.27 Reliability is the ability of a system to produce the same result or outcome in similar populations and situations. Resilience is related to the ability of a complex system to adapt to the changing situations it faces. Ecosystem resilience acknowledges that complex, natural systems cannot be effectively controlled with outcomes that cannot be reliably predicted and is measured by the ability of a system to cope with disruptions before it is forced to adapt.2728Table 4 summarises the three clinical scenarios and the different characteristics for each category of care with respect to CAP.

LIMITATIONS

This report is focused on a single institution’s effort to improve care in an academic medical centre. This effort was spearheaded by a resident physician who had dedicated time, support and mentorship through the Leadership Preventive Medicine Residency Program and the administrative leadership.29 Of interest, after the graduation of the resident leader of the improvement group and cessation of email feedback to providers for each patient admitted with CAP (year 2005), four of the five CAP measures assessed continued to improve over time. Other factors particular to the institution, such as specific resources and administrative attention devoted to quality improvement or initiation of public reporting of performance measures,30 may have been more responsible for or contributed in some way to the observed improvements and may or may not be replicable at other institutions.

The before and after study design cannot exclude the possibility that other local or national changes in the care for CAP patients not described in this report contributed in some way to the improvement in performance that was observed. There is a large body of literature examining the use of complexity science and the study of complex adaptive systems to better understand and improve healthcare systems.2531^–35 There are also concerns that complexity science has been inappropriately applied to healthcare systems and largely misunderstood.36 This paper uses Glouberman and Zimmerman’s typology based on complexity science to better understand the problems or clinical realities that may exist in a particular setting and to design improvement interventions that are matched to that reality. It is beyond the scope of this paper to explore the many other inviting aspects of complexity science such as non-linearity, distributed control, attractors, emergent behaviours and self-organisation, not explicitly addressed in this typology. Although we have presented this typology to characterise clinical realities, we acknowledge that all care processes cannot be categorised into one discrete category and that some situations that have been presented as seemingly simple could actually be complicated or even complex in certain circumstances.

We think that most clinical situations involve all three categories of care. For the case of antibiotic timing, there are aspects that may be simple, complicated and complex. Further, we assume that multiple simple, complicated and complex realities exist in a single patient and for populations of patients. There are also simple situations that may not have known or predictable outcomes and, alternatively, there may be complex situations where the outcome is largely known. We are using the typology to describe commonly occurring features we encountered in the care of our patients with pneumonia. The typology presented in table 4 is not intended to be all-encompassing of every clinical situation that may exist in the field of medicine, rather it is intended as an illustrative guide for the reader interested in exploring the clinical reality they face and to assist in the design of appropriately matched interventions.

INTERPRETATION

Although we have used this typology to “look back” and to help understand the results of our improvement interventions, we believe that this typology might be helpful “looking forward” into the design of future improvement efforts. Our hypothesis is that interventions that are appropriately matched to the clinical reality can increase the success and uptake among providers. This approach to clinical improvement work may offer a new frame within which new conversations might occur and they, in turn, might help clarify the debate on the increasing use of standardisation and the perceived loss of autonomy by some providers. This may also be useful in the development of performance measures and pay for performance initiatives. In the current climate of linking financial incentives to performance measures, it is critical to clarify the circumstances that call for increased measurement and standardisation and those that demand careful consideration of clinical context and patient values. Using the framework proposed in this article, value would be associated with increased standardisation of clinical care in the scenarios on the left side of table 4, and with increased patient and provider autonomy, judgement and flexibility in situations on the right side. The use of such distinctions allows the healthcare system to maximise the investment of the training, skill and qualities of healthcare providers in dealing with complex situations while at the same time attending to the need for increased reliability of the “simpler” aspects of care.

NEXT STEPS

Future work using this typology in a prospective fashion to design and evaluate improvement interventions based on the clinical reality is needed to further develop and validate this model. Based on such prospective work, specific improvement interventions can be identified that are best suited for each of the three different clinical realities. While we have asserted that increasing the standardisation and reliability of simple and complicated care could potentially increase the time for providers to address more complex aspects of patient care, this paper does not provide specific proof. We hope that future studies of improvement work that increases the reliability of simple care will also examine other effects on more complex aspects of care and whether providers had more or less time to devote to adapting care for individual patients. Further development of methods and criteria to identify and classify clinical realities into the different categories are also needed. Empirical studies of intervention efficacy could potentially benefit from clarity about how “matches” were made with the underlying realities of the clinical care.