Symptom evaluation and management is an essential element of clinical care, especially in primary care, where initial evaluation of newly presenting symptoms is a key role and expectation of our patients. While many symptoms resolve spontaneously, others can represent initial manifestation of serious illness. The challenge of initial symptom evaluation is illustrated by studies showing that one-third of patients with common symptoms are not able to be definitively diagnosed, even with intensive investigations.1 Of these, up to half represent unusual symptoms of known chronic illness that is under treatment, but it is often difficult to differentiate these patients prospectively.2 Such variations and uncertainties make it imperative that reliable systems are in place to follow up and monitor symptoms as they evolve or resolve. Such symptom “loop-closing” processes are all the more needed given shifts in care delivery with increasingly diverse types of encounters and communication venues where concerns may be brought to the clinician’s attention.

While the science of diagnosis is well established, there are no well-defined or evidence-based guidelines addressing best practices for reliable and appropriately timed follow-up on new symptoms. Ideally, such symptoms need to be monitored and tracked until they resolve or result in a clear diagnosis, further diagnostics, or treatment. Although healthcare has yet to create perfectly reliable systems to ensure appropriate follow-up on tests and referrals, this issue has received substantial attention.3,4,5,6,7,8,9 Little attention, however, has focused on how to monitor worrisome symptoms over time to ensure they are followed reliably until they resolve or result in a diagnosis.

The standard approach to new symptom evaluation is to perform a focused history and physical exam targeted at most likely or critical diagnoses. A patient presenting with a headache, for example, may describe a visual aura preceding the headache and complain of photophobia which, when coupled with a normal neurologic exam, suggests a diagnosis of a migraine headache. Other patients may have less differentiated headaches but normal results of a neurologic exam, suggesting a benign process that makes initial imaging of low value. If the history and exam are unrevealing, however, a reasonable decision might be to observe the course of the patient’s symptoms at specified future time points. This “watch and wait” strategy may be explicitly, or more often implicitly, chosen. If symptoms improve or resolve, no further evaluation or treatment might be needed, whereas if symptoms persist or worsen and remain unexplained, further monitoring, tests, treatments, or referrals may be indicated.10

Unfortunately, most primary care practices lack defined processes to operationalize this most important test—the test of time. This lack of more fail-safe processes is surprising.11,12,13 In addition to potential adverse health consequences from failure to follow up on symptoms (along with abnormal test results or obtaining and acting on needed referrals), clinicians are at risk for malpractice claims in the event that a patient experiences an adverse outcome associated with poor follow-up. Even the practice of scheduling a patient for a visit to follow up on symptoms is insufficient and inefficient, requiring a practice to use limited time to see a patient whose symptoms have resolved. For patients who do not keep or cancel visits, we often lack systems to close the loop on active concerns.

While failures to “close the loop” on concerning symptoms are an important challenge and vulnerability for busy primary care clinicians, these are exactly the types of process issues that lend themselves to systems engineering analysis and design approaches, increasingly advocated as a complement to other process improvement methods.14,15,16,17,18 The field of systems engineering includes methods for analyzing, designing, and optimizing processes so they perform with high reliability across varied settings and populations. Engineers routinely talk of concepts and terms that are less familiar in healthcare process improvement but offer great potential (e.g., flow simplification, reduction of queues and push systems, non-value-added work, processes that assure loop closing, resilient systems that adapt and operate reliably under routine and novel stresses, and fault-recovery processes that minimize harm when failures occur nonetheless), and the application of systems engineering principles might help quality improvement efforts. As an example drawn from our work: for blood tests ordered during a visit, phlebotomy is embedded in the practice so that patients can visit the lab before they leave the clinic, eliminating wasteful scheduling, travel, and complexity. To smooth flow and reduce queues, patients are triaged to phlebotomists with the shortest waiting time. More than 95% of patients complete blood labs and have results returned to the ordering clinician. For orders originating in telehealth visits, conversely, several additional steps need to occur, dropping completion by 60%. Patients need to be instructed how to obtain lab tests, appointments need to be scheduled, and patients need to travel to a testing facility, all opportunities for completion failures and time delays. Additionally, fault-recovery processes are not in place to reach out to patients and help resolve barriers that interfere with completing tests.

In general, a system engineering approach to the problem of symptom follow-up would start with identifying the functional needs of a reliable symptom follow-up process (i.e., the “what” needs to occur rather than the “how”), followed by generation of multiple potential means for achieving these needs, prototyping and evaluating each, and then detailed design of the most promising ideas. Engineering design principles also would be used to identify and replace low-reliability human-dependent processes with more robust system-based processes (e.g., delegated or automated phone, portal, or texting follow-up) that can also reduce burden and non-value added activities (yellow post-its, phone tag, frustrated patients on hold trying to contact busy understaffed offices).19,20

Examples of engineering methods typically used in such work include systematic analysis and design technique, reliability block diagrams, failure modes and effects analysis, value stream mapping, fault tree analysis, task and cognitive analysis, and computer simulation, to name a few.21 How to standardize workflows, simplify and combine process steps, reduce cognitive and work burden, and incorporate defaults, opt outs, automation, or forcing functions would be considered. It also is likely that achievement of highly reliable processes will necessitate both new process ideas and efforts for “mistake proofing” existing work flow processes. While in some cases this can require more time than typically involved quality improvement activities, results can pay dividends.

The Agency for Healthcare Research and Quality (AHRQ) has sponsored projects to advance patient safety though learning laboratories that foster collaborations between systems engineers and health services quality improvement researchers. Through this process, systems engineers can be helpful in providing new lenses and tools to aid in understanding and redesigning more reliable approaches to common patient safety issues. Our learning laboratory is focusing on closing the loop on patient symptoms (as well as test results and referrals), prioritizing areas of intersection of high clinical risk and high risk of follow-up failure.22,23,24,25 Examples of symptoms warranting reliable follow-up tracking include rectal bleeding, weight loss, chest pain, abdominal pain, skin lesions suspicious for cancer, new headache, or back pain. Each symptom is commonly seen in primary care, generally caused by self-limited or benign conditions, but on rarer occasions can be the harbinger of a serious progressive or life-threatening condition.

Figure 1 displays a simplified clinical conceptual model produced through our work with systems engineers to more reliably follow up on worrisome symptoms. For example, for a patient presenting with new headaches, the history and exam may suggest a diagnosis, or may exclude various worrisome possibilities. Normal results of a neurologic exam may make a brain tumor much less likely, while abnormal exam results may require imaging to evaluate this possibility. When a diagnosis (or differential diagnosis) is made (based on the initial history and exam), we need to ensure the patient’s course and response to treatment is consistent with that diagnostic hypothesis. Tests or referrals may be ordered, which necessitate systems for tracking their completion and follow-up on their results to ensure the loop on that symptom is closed. Testing may lead to confirming a diagnosis or suggesting an alternate diagnosis (and a possible alternate treatment) or might be fully normal and reassuring.

Figure 1
figure 1

Symptom surveillance automated monitoring to operationalize test of time, showing initial symptom awareness, decisions to monitor, ordered labs, referrals, and steps in closing each step.

In cases where no tests or referrals are indicated based on initial evaluation, for example, a patient with an exacerbation of headaches or low back pain without any “red flag” symptoms that usually resolves with time, monitoring the symptoms (the “test of time”) would be an appropriate plan. For example, when benign self-limited nonspecific headache or low back strain is the most likely diagnosis, monitoring is reasonable but only if reliably “hard-wired” follow-up mechanisms are in place to recognize if any new red flags develop or that patient’s symptoms adequately resolve and are not the presenting symptoms of a more worrisome diagnosis.

Ideally, such automated systems should leverage the potential of the electronic medical record (EMR) to flag and track completion of ordered tests and referrals, as well as to implement a process to complete incomplete referrals and tests. EMRs also need the capacity to store and track patients with worrisome symptoms that need follow-up. Although both EMR innovation and re-engineering of office staff workflows would be required, the benefits would be obvious and measurable for clinicians and patients.

Of course, few newly redesigned processes work perfectly from the start. Each aspect needs to be piloted in multiple settings, refined based on lessons learned, or replaced with alternate ideas, thereby over time progressing from current “1 sigma” (or worse) to “6 sigma” process reliability.26 The design of a reliable system requires active input from patients, clinicians, and staff, ideally with the help of IT and systems engineers to improve overall performance and efficiencies. Process details and features need to be modified based on local resources, workflows, and patient preferences.

In other health systems, other approaches might be taken to achieve reliable closed-loop symptom monitoring. As a roadmap, Figure 2 depicts a common system engineering process design method for a symptom monitoring process, as a precursor to identifying specific potential solutions to each functional need. These functions logically would include processes for designating and flagging a concerning symptom for follow-up, assigning an appropriate timeframe, instituting and activating mechanisms for monitoring a patient’s symptom status, and determining and documenting subsequent diagnosis or care decisions, each with needed inputs, desired outputs, enabling mechanisms, and governing constraints (the arrows in the figure).

Figure 2
figure 2

Engineering design of a systematic process for following up on worrisome symptoms. Hierarchical structured analysis and design technique (SADT) diagram showing functional needs, inputs, outputs, mechanisms, and constraints.

If symptoms are worrisome to the clinician or patient, they warrant reliable follow-up to ensure they resolve or are appropriately followed up. Fail-safe systems do not assume “no news is good news” but rather have proactive mechanisms to ensure symptoms are followed until they are sufficiently explained, resolved, reassessed, or treated. Given current stresses on busy clinicians and staff, care discontinuities, communication challenges, dis-coordinated handoffs, and suboptimal information systems, efficient systems are critical as an alternative to adding yet more burden. Such a transformation to processes to ensure higher reliability without additional work needs to be innovatively imagined, expertly designed, and implemented. The impact of reliable systems to monitor and “close the loop” on worrisome symptoms is likely to reduce worry for primary care clinicians while improving diagnosis, appropriate testing, referrals and treatments, patient outcomes, and malpractice risk. Symptom tracking mechanisms may enable learning systems to gather data on the likelihood that symptoms tracked will result in the discovery of serious illness and inform how an earlier evaluation might diagnose these problems sooner.