Difficult diagnoses in the ICU

The majority of patients admitted to the ICU have common diseases. The presentation may be severe, but the pathology is bread-and-butter: sepsis, heart failure, pneumonia, GI bleeding, and so on.

However, a certain number of patients present without a clear underlying diagnosis. Although they have some overt physiologic abnormality requiring support in the ICU, such as hypotension or obtundation, its etiology is not obvious. Indeed, they may go their entire ICU course without such a diagnosis being made; often, non-specific supportive care seems to be enough. The patient improves and goes on their way. What happened? Who knows.

These cases can be vexing. More vexing, though, are the cases that don’t improve; they die or suffer significant morbidity. How many of these could have been prevented had a diagnosis been made? This is an unknown quantity and nobody can say. Even the cases that do recover in the absence of a diagnosis are a bitter pill: intellectually unsatisfying at the best, and who’s to say they might not have recovered more robustly (faster or to a better baseline) with a proper diagnosis?

In truth, such cases are not that uncommon. Whether it’s because modern patients are complicated, or because the setting of critical illness makes diagnosis difficult, or simply because we’re all busy and it may never seem like one’s job or the right moment to spend hours meditating over rare diseases, we tend to move through our shifts without trying very hard to differentiate the undifferentiated. As a result, the diagnoses slip through our fingers like sand through a surgical sieve.

Here are a few thoughts on making challenging ICU diagnoses.

Basic principles of difficult diagnosis

  1. One must consider a diagnosis in order to make it, but none of us routinely considers uncommon things. This is an inevitable consequence of human psychology and is not fixable. So to make these diagnoses, the ordinary workflow is inadequate; we must actively set down one tool (our everyday, routine diagnostic mode using Kahneman’s System 1 thinking) and activate a different, more rigorous and thoughtful one (System 2). The first step to making a difficult diagnosis is the explicit decision to try.
  2. The most common diagnoses will always be common diseases. It is not possible to determine a priori whether a patient has a common or uncommon disease. For both reasons, one must rule out common problems before considering the uncommon. After this has been done, and either the patient has failed to improve, or elements of the presentation still fail to “fit the story,” less common diagnoses should be considered.
    • Corollary: From the clinician’s perspective, diseases tend to present—and should be considered—in the following order of frequency:
      1. Common presentations of common diseases
      2. Uncommon presentations of common diseases
      3. Diseases falsely believed to be uncommon, but actually not
      4. Truly uncommon diseases
  3. Overall, it may be plausible that a patient has some uncommon disease. However, a priori it is usually unlikely that they have any particular uncommon disease. This highlights the need for a rational approach, because there are too many diseases to reach the answer by random guesswork or shotgun testing (the elimination approach); the numbers are against you.
  4. By the time you realize you don’t know what’s on, the accumulated, largely inaccurate working diagnoses are generally more harmful than helpful. Erase them and start over. “Readmit” the patient with fresh eyes, obtaining a new history and physical exam, and assuming nothing.
    • Corollary: In the complex modern patient, a key component of the history of most problems is the documentation of recent inpatient and outpatient care. This is often lengthy and dense and may be time-costly to unpack, but is always worthwhile.
  5. The most important diagnoses to make:
    • Are treatable (or are untreatable, but the diagnosis has a major impact on prognosis; or the diagnosis would obviate the need for morbid treatment of other misdiagnosed conditions, including multiple readmissions for flares of an unknown disease)
    • Require specific treatment, which would not otherwise be offered
    • Portend high morbidity or mortality
  6. Keep pulling the threa. Pursue the diagnosis to the end rather than stopping at empiric treatment; a confirmed diagnosis has significant downstream advantages over a mere hypothesis, even when it does not change the immediate care.
    • Corollary: Follow Sutton’s Law. If a certain test will definitively confirm or disconfirm a plausible diagnosis, get it. These cases tend to involve a journey with multiple cycles of “hypothesis-test-repeat.” Too many suspected, possible, or probable working diagnoses result in accumulating uncertainty with accumulating room for error. Better to get the test, even if relatively invasive or inconvenient (such as a biopsy, angiogram, or EMG), than to continue building a diagnostic ladder on an increasingly wobbly foundation.
  7. These cases are plagued by a lack of objectivity. Biases are common, either from the desire for efficiency (an obstacle to diagnosis), the opposing desire for an “interesting case” (an obstacle to correct diagnosis), or diagnostic anchoring (which leads clinicians to adhere to their initial impressions). These cases also tend to regress teams to an “oracular” model of medical care, where the chance of a diagnosis, or even the amount of effort towards it, depends entirely on the personalities, knowledge, and proclivities of the rotating staff who happen to care for the patient. A systematic approach limits these problems.
  8. Do not ignore patently abnormal findings. Allow new information to generate new hypotheses instead of being forced to fit old hypotheses.
    • Caveat: if an abnormal finding can be explained by an already-existing diagnosis, it is more likely to be caused by that. However, this is not true for new findings (new to the patient, not merely to the clinician), which are less likely to be caused by old problems.
  9. Blind “shotgun” consultation of specialists will almost never yield the answer; it just transfers the same challenges to a new set of brains that lack the same familiarity or investment in the case. Don’t count on consultants, or on downstream follow-up by hospitalists or outpatient providers, by which point the patient will have acquired diagnostic inertia. In general, if these diagnoses are to be made, they’ll be made by the primary team. Consultants are most useful to investigate and confirm a diagnosis once it has been proposed.

Common characteristics of missed diagnoses in the ICU population

  1. The disease in question must be capable of causing critical illness, or tend to coexist with diseases that do. This limits the differential significantly.
  2. Since ICU patients routinely receive an array of broad diagnostic tests, for a diagnosis to be missed, it should not be readily detectable on routine lab studies, or its manifestations should be transient or non-specific.
  3. It should generally not be structural, or its appearance should be very non-specific, as such lesions tend to appear on routine imaging. (An exception might be vascular lesions, as angiography is not always routinely performed.)
  4. Unless your ICU population is pediatric, the disease should not be congenital or limited to the young. (An exception is the occasional congenital condition that presents in adulthood.)
  5. It should generally not be limited to special populations, such as the obviously immunosuppressed; these patients are red flags and tend to receive appropriate attention by specialists.
  6. Most modern patients have multiple problems, and most are not mutually exclusive. When considering why a patient might have a second problem in addition to an already-diagnosed one, one should ask “why?” (the Occam’s Razor test), but also acknowledge that the answer may be simply “why not?” Coincidental timing is often explainable by the fact that chronic conditions can be unmasked by acute illness.

A systematic approach to the diagnostic dilemma

  1. Rule out common/ordinary diseases.
  2. Recognize the potential presence of an unusual diagnosis.
  3. “Reboot” the admission from the beginning, ignoring the existing differential, with the goal of objective data collection. Start with gathering a fresh, complete history of the present illness, with an emphasis on:
    • Symptoms and events prior to the acute hospitalization
    • A complete, thorough review of the current hospitalization and other recent care, including tests performed, transient signs and symptoms, and therapeutic trials.
    • Review of laboratory and imaging tests, including abnormal findings previously considered incidental or insignificant.
  4. Gather background information, with an emphasis on:
    • A complete medical and surgical history, including any complications
    • Social history, including occupational, home, and recreational environments
    • Potential toxins, including supplements, polypharmacy, and recreational drugs
    • Family history, focusing on obvious heritable disorders or on hypotheses generated by the patient’s symptoms
  5. Perform a thorough review of systems, with an emphasis on constitutional and neurologic symptoms.
  6. Perform a thorough physical examination, with emphasis on:
    • A complete examination of the skin with attention to any lesions and to easily missed areas (such as the back, skin folds, perineum, and soles of the feet)
    • The mouth and mucous membranes
    • The eyes (potentially including a fundoscopic exam)
    • The lymph nodes
    • A detailed neurologic exam
  7. Build a new differential and order appropriate confirmatory tests. Accept that relatively broad testing will be necessary, but also that broad, targeted testing is distinct from blind or “shotgun” testing. Accept that reaching a diagnosis will likely require a process of testing and development of the differential, rather than a sudden “Eureka!” moment.
  8. Consult a checklist of easy-to-miss diagnoses to see if any might be relevant, such as the one below.

A checklist of easily missed ICU diagnoses

This is an early attempt at creating such a list; please feel free to submit your suggestions. It roughly follows the above principles of conditions that are important to diagnose while also being easy to miss. Conditions in bold may develop de novo during hospitalization.

Difficult diagnoses in the ICU generally start with a syndrome, not with a clear mechanism. The most common phenotypes are undifferentiated encephalopathy, unexplained neuromuscular weakness, suspected but undiagnosed infection (e.g. fever or leukocytosis of unknown origin), and refractory shock or organ failure. The list below is organized by system only for convenience.


These are usually causes of encephalopathy, odd focal deficits, or neuromuscular weakness

  • Demyelinating disorders such as Guillain-Barré syndrome
  • Myasthenia gravis
  • Lambert Eaton myasthenic syndrome
  • Catatonia
  • Anti-NMDA and other autoimmune forms of encephalitis
  • Wernicke encephalopathy
  • Multiple sclerosis
  • Partial (especially temporal) or subclinical seizures
  • Periodic paralyses

These tend to be causes of refractory shock

  • Left ventricular outflow tract obstruction (e.g. from HOCM, SAM)
  • Pulmonary hypertension and RV failure
  • Abdominal compartment syndrome
  • Idiopathic capillary leak syndrome
  • Air embolism
  • Normotensive hypoperfusion (inadequate organ flow despite normal pressure, such as in the setting of cardiogenic shock, kidneys adapted to chronic hypertension, or pressure-dependent stenoses in the cerebral vasculature)

These are often cryptogenic causes of difficulty liberating from the ventilator

  • Vocal cord dysfunction (paradoxical vocal cord motion)
  • Obscure pneumonias, e.g. eosinophilic, COP, etc.
  • Severe OHS/OSA
  • Diaphragmatic weakness
  • Anatomic ateriovenous shunts, such as PFO or pulmonary or hepatic AVMs (usually causing unexplained hypoxemia)

Often associated with liver failure or miscellanous metabolic or neurologic findings

  • Hemochromatosis
  • Vitamin/mineral deficiency (most often in alcoholics, the severely malnourished, or those with absorption problems such as small intestinal bacterial overgrowth)
  • Wilson’s disease
  • Refeeding syndrome
  • Hepatic encephalopathy

Usually causing unexplained renal failure

  • Obstructive nephropathy
  • Acute interstitial nephritis

Usually these are causes of vague toxic syndromes that may be confused with sepsis

  • Hemophagocytic lymphohistiocytosis (HLH)
  • Thrombotic thrombocytopenic purpura (TTP)
  • Intravascular (e.g. large B-cell) lymphoma

Protean in manifestation

  • Myxedema
  • Thyrotoxicosis
  • Adrenal insufficiency (often considered, but still often missed)

Usually causing fevers or outright sepsis with negative cultures and no response to typical broad-spectrum antibiotics

  • Zoonotic diseases, including tickborne illnesses
  • “Exotic” organisms endemic to international settings but not the local environment
  • Tuberculosis, including disseminated TB without clear pulmonary lesions
  • Fungal infections
  • Pre-existing, undiagnosed viral hepatitis or HIV
  • Occult abscesses or other nidus of infection (e.g. vaginal or rectal sources, septic joints, etc)
  • Septic thrombophlebitis
  • Vibro vulnificus infection
  • Sinusitis
  • Atypical pneumonias (mycobacterial, Legionnaire’s, etc)
  • Whipple disease

Protean in manifestation

  • Various systemic autoimmune conditions, such as lupus, amyloidosis, etc.

Often causing vague/bizarre neurologic or metabolic findings

  • Unrecognized recreational drug use
  • Serotonin syndrome or neuroleptic malignant syndrome
  • Propofol infusion syndrome
  • Heavy metal toxicity
  • Cyanide or carbon monoxide toxicity
  • Methemoglobinemia
  • Vitamin/mineral overload
  • Misplaced, migrated, extravasated, or eroded catheters or devices
  • Adverse drug effects
  • Newly emerging infectious disease, such as HIV or COVID-19
  • Novel toxicologic syndromes, such as vaping-associated lung injury or designer drugs
  • Closed-angle glaucoma
  • Familial Mediterranean Fever
  • Acute intermittent porphyria
  • Factitious or conversion disorders (although these are probably more likely to be overdiagnosed than missed)


This framework is evolving, and suggestions are welcome. In the end, the key to making a tough diagnosis is to try—really, truly try—and to follow the diagnostic train until it reaches its destination, rather than giving up at some intermediate stop and being content to offer mere supportive care. Most providers know how to diagnose, although they may find it to be an atrophied skill; diagnosis arises from an excellent history, a thorough physical, some careful thought, and a series of rational tests and therapeutic trials. That’s the process of diagnostic medicine, as it always has been, and it holds true even in the high-stakes setting of the ICU.