on 21-Jul-2022 (Thu)

In this paper, we offer marketing analysts an alternative to these models by developing a deep learning based approach that does not rely on any ex-ante data labelling or feature engineering, but instead automatically detects behavioral dynamics like seasonality or changes in inter-event timing patterns by learning directly from the prior transaction history

To circumvent additional feature engineering when increasing model flexibility, Salehinejad and Rahnamayan (2016) and Mena, Caigny, Coussement, Bock, and Lessmann (2019) have introduced a recurrent neural network (RNN) 2 approach to the domain of customer base analysis by modeling the evolution of RFM variables over time. However, since the focus still remains on predicting hand-engineered RFM metrics, such an approach does not fully leverage the automatic feature extraction capabilities of deep learning methods. Sheil, Rana, and Reilly (2018) take this one step further by allowing the neural network to derive its own internal representation of transaction histories. The authors demonstrate the performance of several RNN architectures and benchmark them against more conventional machine learning approaches for predicting purchasing intent. In a similar context, Toth, Tan, Di Fabbrizio, and Datta (2017) have shown that a mixture of RNNs can approximate several complex functions simultaneously. More recently, Sarkar and De Bruyn (2021) demonstrate that a special RNN type can help marketing response modelers to benefit from the multitude of inter-temporal customer-firm interactions accompanying observed transaction flows for predicting the most likely next customer action. However, their approach is limited to single point, next-step predictions and to continue with such forecasts into the long-run one must estimate the new model repeatedly with each additional future time step

Appendix B.

Technical Implementation Notes

Here we provide more details about the technical aspects of the model and share our experience with training. We implement the model in Python using the open source neural network tools Keras (Chollet, 2015) and Tensorflow (Abadi et al., 2015) and we use a standard desktop computer with an NVidia GeForce 2080 consumer graphics card for training. This hardware setup allows for multiple (typically 4–8, depending on model and data size) LSTM models to train in parallel, so lesser hardware is fine to use. We tried replacing the LSTM memory layer with different and more advanced RNN architectures, but the ‘‘vanilla” LSTM speeds up training by a factor of 10 due to its efficient GPU-acceleration, which is an advantage that outweights any other potential incremental benefits. If less granular forecasts are sufficient, an additional training speed-up can be achieved by aggregating the input samples into monthly rather than weekly buckets. As illustrated for the Charity Contributions case (see the second entry in Table 4) this does not result in a significant loss of accuracy. We observe the same findings for different sub-groups of customers and variations in holdout periods

Sheil, Rana, and Reilly (2018) take this one step further by allowing the neural network to derive its own internal representation of transaction histories. The authors demonstrate the performance of several RNN architectures and benchmark them against more conventional machine learning approaches for predicting purchasing intent.

However, many patients, including those with large PE, have mild or nonspecific symptoms or are asymptomatic. For example, a meta-analysis of 19 studies (25,343 patients) found that clinical impression alone had a sensitivity and specificity of 85 and 51 percent, respectively, for the diagnosis of PE [ 9].Thus, it is critical that a high level of suspicion be maintained such that clinically relevant cases are not missed.

The most common symptoms in patients with PE were identified in the Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED) group (table 1) [6]. They include the following:

● Dyspnea at rest or with exertion (73 percent)

● Pleuritic pain (66 percent)

● Cough (37 percent)

● Orthopnea (28 percent)

● Calf or thigh pain and/or swelling (44 percent)

● Wheezing (21 percent)

● Hemoptysis (13 percent)

Less common presentations include transient or persistent arrhythmias (eg, atrial fibrillation), presyncope, syncope, and hemodynamic collapse (<10 percent each) [10,11]. Hoarseness from a dilated pulmonary artery is a rare presentation (Ortner syndrome) [12].

The onset of dyspnea is frequently (but not always) rapid, usually within seconds (46 percent) or minutes (26 percent) [8]. Dyspnea may be less frequent in older patients with no previous cardiopulmonary disease.

Dyspnea is more likely to be present in patients who present with PE in the main or lobar vessels.

Some patients have a delayed presentation over weeks or days. One prospective study reported that patients with a delayed presentation beyond one week tended to have larger, more centrally located PE compared with patients who presented within seven days (41 versus 26 percent) [22 ]

Although the true incidence of asymptomatic PE is unknown, one systematic review of 28 studies found that, among the 5233 patients who had a deep vein thrombosis (DVT), one-third also had asymptomatic PE [23].

Common presenting signs on examination include [8]:

● Tachypnea (54 percent)

● Calf or thigh swelling, erythema, edema, tenderness, palpable cords (47 percent)

● Tachycardia (24 percent)

● Rales (18 percent)

● Decreased breath sounds (17 percent)

● An accentuated pulmonic component of the second heart sound (15 percent)

● Jugular venous distension (14 percent)

● Fever, mimicking pneumonia (3 percent)

PE is a common cause of sudden cardiac arrest or circulatory collapse (8 percent), especially among patients younger than 65 years old [8,24,25 ]. Among such patients, either dyspnea or tachypnea is present in 91 percent.

Massive PE may be accompanied by acute right ventricular failure manifested by increased jugular venous pressure, a right-sided third heart sound, a parasternal lift, cyanosis, and obstructive shock. However, shock may also develop in patients with smaller PE who have severe underlying pulmonary hypertension.

PE should be suspected anytime there is hypotension accompanied by an elevated central venous pressure that is not otherwise explained by acute myocardial infarction, tension pneumothorax, pericardial tamponade, or a new arrhythmia [ 26,27].

Unexplained hypoxemia in the setting of a normal chest radiograph should raise the clinical suspicion for PE and prompt further evaluation.

However, while often abnormal among patients suspected of having PE, ABGs can be normal in up to 18 percent of patients with PE [ 28].

Common abnormalities seen on ABGs include one or more of the following [6,28,30] (see "Arterial blood gases"):

• Hypoxemia (74 percent)

• Widened alveolar-arterial gradient for oxygen (62 to 86 percent)

• Respiratory alkalosis and hypocapnia (41 percent)

Abnormal oxygenation may be of prognostic value. As an example, patients with hypoxemia or room air pulse oximetry readings <95 percent at the time of diagnosis are at increased risk of complications, including respiratory failure, obstructive shock, and death [31].

As markers of right ventricular dysfunction, troponin levels are elevated in 30 to 50 percent of patients who have a moderate to large PE [34,39]

Troponin elevations usually resolve within 40 hours following PE, in contrast to the more prolonged elevation after acute myocardial injury [ 40].

The most common findings are tachycardia and nonspecific ST-segment and T-wave changes (70 percent) [7].

Abnormalities historically considered to be suggestive of PE (S1Q3T3 pattern, right ventricular strain, new incomplete right bundle branch block) are uncommon (less than 10 percent) [46,47].

ECG abnormalities that are associated with a poor prognosis in patients diagnosed with PE include [ 41,42,44]:

● Atrial arrhythmias (eg, atrial fibrillation)

● Bradycardia (<50 beats per minute) or tachycardia (>100 beats per minute)

● New right bundle branch block

● Inferior Q-waves (leads II, III, and aVF)

● Anterior ST-segment changes and T-wave inversion

● S1Q3T3 pattern

Chest radiograph — Nonspecific abnormalities on chest radiography are common (eg, atelectasis, effusion) in PE, but a normal chest radiograph can be seen in 12 to 22 percent of patients [6,7,48].

However, it [chest radiograph] is not necessary if a CTPA is planned.

Hampton's hump is a shallow, hump-shaped opacity in the periphery of the lung, with its base against the pleural surface and hump towards the hilum (image 1).

Westermark's sign is the demonstration of a sharp cut-off of pulmonary vessels with distal hypoperfusion in a segmental distribution within the lung ( image 2).

Several approaches for hemodynamically stable nonpregnant adult patients with suspected PE have been proposed [1-3,56-62]. Their purpose is to efficiently diagnose all clinically important PE while simultaneously avoiding the risks of unnecessary testing. We prefer an approach that selectively integrates clinical evaluation, three-tiered pretest probability (PTP) assessment, PE rule out criteria (PERC), D-dimer testing, and imaging (algorithm 1 and algorithm 2 and algorithm 3).

Empiric anticoagulation while waiting for test results should be individualized according to the clinical suspicion for PE, the anticipated timing of definitive testing, and the risk of bleeding, the details of which are discussed separately. (See "Treatment, prognosis, and follow-up of acute pulmonary embolism in adults", section on 'Empiric anticoagulation' and "Venous thromboembolism: Initiation of anticoagulation".)

. Although gestalt estimates and calculating probability scores have comparable sensitivity when combined with D-dimer testing, meta-analyses suggest that probability scores may have higher specificity [9,68] and increase the diagnostic yield of CTPA [69].

Wells criteria have best validated in outpatients presenting with suspected PE.

However, one study of hospitalized patients, reported a sensitivity and specificity of 72 and 62 percent, respectively [ 73]; the addition of D-dimer to Wells criteria improved the sensitivity to 99 and reduced the specificity to 11 percent.

A retrospective study compared the Wells score with the YEARS algorithm and found that the YEARS algorithm was more sensitive (97 versus 74 percent) but less specific (14 versus 34 percent) for the diagnosis of PE [74]. The YEARS algorithm is described below. (See 'D-dimer' below.)

For patients with a low probability of PE (eg, PTP <15 percent, Wells score <2), we apply the PERC (table 4) to determine whether or not diagnostic evaluation with D-dimer is indicated (algorithm 3).

When the PERC rule is chosen, the following applies to patients for whom the clinician has determined that a diagnostic evaluation for PE is indicated (see 'PERC rule' below):

● For patients who fulfill all eight PERC criteria, no further testing is required

● For patients who do not fulfill all eight criteria, further testing with sensitive D-dimer measurement is indicated

In low-risk patients where PERC cannot be applied (eg, inpatients, critically-ill patients) or PERC is positive, D-dimer testing is indicated and the following applies (see 'D-dimer' below):

● When the D-dimer level is <500 ng/mL (fibrinogen equivalent units), no further testing is required

● When the D-dimer level is ≥500 ng/mL (fibrinogen equivalent units), diagnostic imaging should be performed, preferably with CTPA

The PE rule out criteria (PERC) rule was designed to identify patients with a low clinical probability of PE in whom the risk of unnecessary testing outweighs the risk of PE [2,75-77] (see 'Low probability of pulmonary embolism' above). The PERC rule has eight criteria (table 4):

• Age <50 years

• Heart rate <100 beats/minute

• Oxyhemoglobin saturation ≥95 percent

• No hemoptysis

• No estrogen use

• No prior DVT or PE

• No unilateral leg swelling

• No surgery/trauma requiring hospitalization within the prior four weeks

PERC is only valid in clinical settings (typically the ED) with a low prevalence of PE (<15 percent) [76]. In clinical settings with a higher prevalence of PE (>15 percent), the PERC-based approach has been shown to have a substantially poorer predictive value [76]. Thus, it should not be used in patients with an intermediate or high suspicion for PE or for inpatients suspected as having PE.

For patients in whom the risk of PE is thought to be low, a normal D-dimer (<500 ng/mL [fibrinogen equivalent units]) effectively excludes PE, and typically no further testing is required. This includes patients who have had a prior PE and those with a delayed presentation and hospitalized patients [22,73,80]

For most patients in whom the risk of PE is thought to be intermediate, a normal D-dimer (<500 ng/mL [fibrinogen equivalent units]) also effectively excludes PE, and typically no further testing is required. However, some experts believe that a subset of patients in the intermediate risk category (eg, those in the upper zone of the intermediate range [eg, Wells score 4 to 6 or Modified Geneva sore 8 to 10] or patients with limited cardiopulmonary reserve) should undergo imaging based upon the higher probability of PE in these patients since the sensitivity of D-dimer is not as good.

While a negative D-dimer result does reduce the likelihood of PE in this population, it does not reduce it sufficiently to rule out the diagnosis, with some data suggesting a prevalence of PE of 5 percent or more in those with a high pretest probability and a negative D-dimer [81-85]

While D-dimer assays are highly sensitive, their specificity is low, usually between 40 and 60 percent. D-dimer results are often falsely positive, and the proportion of false positive results increases with certain clinical conditions and any acute or inflammatory process (eg, age >50 years, recent surgery or trauma, acute illness, pregnancy or postpartum state, rheumatologic disease, renal dysfunction [estimated glomerular filtration rate <60 mL/min/1.73 m²]), and sickle cell disease (table 5) [26,88-92].

Adjusted D-dimer — Although high-sensitivity D-dimer testing is preferred, adjusted D-dimer levels based on certain criteria have been proposed and may be considered as an alternative in patients with a low probability or low intermediate probability for PE. They should not be used in those with high-probability or intermediate-high-probability for PE.

D-dimer levels rise with age such that using the traditional cutoff value of <500 ng/mL (fibrogen equivalent units) results in reduced specificity of D-dimer testing in older patients (>50 years), a population in whom PE is common. Several studies report its use [2,61,93-98] with the most commonly used formula for age adjustment as:

Age (if over 50 years) x 10 = cutoff value in ng/mL (fibrinogen equivalent units)

Alternate D-dimer cut-offs have also been used. In one prospective multicenter study, 3465 patients with suspected PE from an outpatient setting underwent D-dimer testing and an assessment of pretest probability using the YEARS criteria [99].

The YEARS criteria include three items from the Wells score: clinical signs of DVT, hemoptysis, and PE as the most likely diagnosis, all scored as a yes or no. PE was excluded in patients with zero YEARS items and a D-dimer level <1000 ng/mL, and patients with ≥1 YEARS item and a D-dimer <500 ng/mL.

Intermediate probability of pulmonary embolism — For most patients in whom the suspicion for PE is intermediate, a sensitive D-dimer level should be measured (algorithm 2). (See 'D-dimer' above.)

● When the D-dimer level is <500 ng/mL (fibrinogen equivalent units), no further testing is typically required. However, some experts will proceed with diagnostic imaging in select patients. For example, imaging may be considered in patients who have limited cardiopulmonary reserve (ie, patients in whom PE would not be well tolerated) or those in whom the clinical probability of PE was in the upper zone of the intermediate range (eg, a Wells score of 4 to 6 or a Geneva score of 8 to 10).

● When the D-dimer level is ≥500 ng/mL (fibrinogen equivalent units), diagnostic imaging should be performed, preferably with CTPA.

When imaging is indicated, CTPA is the imaging modality of choice. V/Q scan is reserved for patients in whom the CTPA is contraindicated (eg, history of moderate or severe contrast allergy, high risk of contrast nephropathy [estimated glomerular filtration rate <30 mL/min/1.73 m²], or inability to tolerate computed tomography [CT] scanning due to class 2 or 3 obesity or difficulty lying flat).

V/Q may also be indicated when CTPA is inconclusive, or when additional testing is needed, such as when the clinical suspicion of PE remains high despite negative imaging. Hypotension and advanced heart failure may limit the circulation of IV contrast and increase the likelihood of an indeterminate CTPA [ 106].

V/Q scan – For patients in whom a V/Q scan is performed, management is dependent upon the interpretation of the scan in the context of the pretest clinical probability for PE. Although the Wells criteria were developed after the Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED) study [5], it is appropriate to use Wells to stratify risk (table 2 and table 6) for the purposes of interpretation

• In patients with a normal V/Q scan and any clinical probability, no further testing is necessary.

• In patients with a low-probability V/Q scan and low clinical probability (eg, Wells score <2 (table 2) (calculator 1)), no further testing is necessary.

• In patients with a high-probability V/Q scan and high clinical probability (eg, Wells score >6 (table 2) (calculator 1)), immediate treatment is indicated. (See "Treatment, prognosis, and follow-up of acute pulmonary embolism in adults".)

• All other combinations of V/Q scan results and clinical pretest probabilities are indeterminate (inconclusive), and further testing is required.

In some cases, if contraindications to CTPA are present but can be readily resolved (eg, premedication for a contrast allergy) and alternate imaging such as V/Q scanning is not feasible, CTPA may be performed after a short delay (eg, 8 to 12 hours) with consideration for empiric anticoagulation while waiting.

For optimal image quality, the patient should be able to hold still and hold their breath for about 30 seconds.

A chest CT with contrast not performed as a CTPA but for other indications may incidentally detect pulmonary emboli but is not an adequate exam for excluding suspected PE [107].

Lower-extremity ultrasound with Doppler — A new diagnosis of DVT in the setting of symptoms consistent with PE is highly suggestive, although not definitively diagnostic, of PE. As such, Doppler ultrasonography can be used as an initial test in the evaluation of suspected PE, as positive results can justify initiating anticoagulant treatment. However, because of the low sensitivity of Doppler ultrasonography in this setting [132,133], it is not sufficient to rule out PE and may be most useful for patients suspected of having a PE but in whom definitive imaging (eg, CTPA, V/Q scanning) is contraindicated, indeterminate, or likely to be delayed.

Although echocardiography has limited value diagnostically, it is most useful for prognostic purposes in patients with confirmed PE (eg, new RV strain and RV thrombus are poor prognostic indicators), the details of which are discussed separately.

Approximately 30 to 40 percent of patients with PE have echocardiographic abnormalities indicative of RV strain or pressure overload [155-157] and data suggest that there is direct correlation between the extent of RV dysfunction and the degree of perfusion defects on lung scans [152,153]. RV findings include:

● Increased RV size

● Decreased RV function

● Tricuspid regurgitation

● Abnormal septal wall motion

● McConnell's sign

● Decreased tricuspid annular plane systolic excursion (TAPSE)

Regional wall motion abnormalities that spare the right ventricular apex (McConnell's sign) are insensitive (77 percent) for the diagnosis of PE but, in those who demonstrate this sign, it may be used to distinguish patients with RV strain from acute PE from those with pulmonary hypertension, who tend to have global RV dysfunction [158].

In a cohort of 79 patients with acute PE receiving anticoagulant therapy, complete clot resolution occurred in 40 percent of patients within one week, 50 percent within two weeks, 73 percent within four weeks, and 81 percent by four weeks or longer [169]. Resolution was quicker in larger (main and lobar) pulmonary arteries compared with smaller (segmental and subsegmental) vessels, particularly during the first week.

Ventilation perfusion (V/Q) scanning – A segmental or subsegmental perfusion defect with normal ventilation are diagnostic of PE. Images are interpreted as high, intermediate, or low probability of PE or normal. A normal scan and a low probability scan in the setting of low clinical probability of PE are sufficient to exclude PE. A high-probability V/Q scan and high probability of PE confirms PE. All other combinations of V/Q results and clinical probability are nondiagnostic.

Computed tomographic pulmonary angiography (CTPA) or magnetic resonance pulmonary angiography (MRPA) – A filling defect in any branch of the pulmonary artery (main, lobar, segmental, subsegmental) that becomes evident after contrast enhancement is diagnostic of PE (image 5 and image 6) [115]. Indeterminate or nondiagnostic scans are reported when a filling defect is not clearly visualized (eg, embolus in a small peripheral pulmonary artery, poor contrast enhancement, image degradation by motion or metallic beam hardening artefact).

Echocardiography is rarely diagnostic of PE but a presumptive diagnosis may be made in patients who are hemodynamically unstable so that life-saving therapy can be administered.

Lower-extremity proximal vein compressive ultrasound (US) demonstrating deep venous thrombosis (DVT) is not diagnostic of PE but can be used to justify treatment in emergent settings or when other testing cannot be obtained.

For patients who present with signs and symptoms of PE, the major competing diagnoses include heart failure, pneumonia, myocardial ischemia or infarction, pericarditis, acute exacerbations of chronic lung disease, pneumothorax, and musculoskeletal pain.

Dyspnea – Dyspnea that is abrupt in onset or disproportionate to the patient's underlying lung function, or dyspnea that occurs with hypoxemia, hemoptysis, and/or pleuritic chest pain may favor a diagnosis of PE.

Chest pain – Acute chest pain, especially pain that is pleuritic in nature, is highly suspicious for PE, but may also be due to other etiologies such as pneumonia, pericarditis, pleuritis, and rib fracture.

Hemoptysis – Hemoptysis that occurs with pleuritic pain and hypoxemia should prompt consideration of acute PE, but can also be secondary to pneumonia or heart failure (often frothy and pink).

Leg pain and swelling – Unilateral leg swelling should raise the suspicion for PE in association with deep vein thrombosis (DVT), while bilateral swelling may be more supportive of heart failure.

Syncope – Syncope in patients without a clear precipitant should raise suspicion for PE [18].

Hypoxemia – Hypoxemia (partial pressure of oxygen in arterial blood on room air <80 mmHg [10 kPa]) in the setting of a normal chest radiograph, or hypoxemia that is disproportionate to the chest radiograph appearance, should prompt consideration of PE as well as the following alternate diagnoses

Tachycardia – Unexplained tachycardia, especially in a patient with risk factors for PE, should prompt clinicians to consider PE, but can be associated with other diagnoses, including cardiac arrythmia, sepsis, hypovolemia, drugs, and toxins.

Heart failure – The combination of dyspnea and leg swelling due to heart failure may mimic PE. Evidence of pulmonary edema may be supported by crackles and chest radiography. While brain natriuretic peptide elevation can support heart failure, this can also be seen in acute PE.

Pneumonia – Fever, consolidation on chest imaging, and leukocytosis may favor infection over PE, but can also be the presenting features of an acute lobar pulmonary infarct secondary to PE, particularly as it evolves over the first few days or weeks. The presence of risk factors for PE, persisting symptoms or poor response to antibiotics, or abrupt onset of new symptoms during the course of subacute illness should prompt the clinician to investigate for PE. (See "Clinical evaluation and diagnostic testing for community-acquired pneumonia in adults" and "Epidemiology, pathogenesis, and microbiology of community-acquired pneumonia in adults" and "Nonresolving pneumonia".)

Pericarditis – The pain of pericarditis can be pleuritic and therefore mimic PE. The presence of a viral prodrome, pre-existing inflammatory disease, and electrocardiographic findings of ST elevation may increase the likelihood of pericarditis.

Exacerbation of underlying chronic lung disease – Patients with chronic lung disease often present with dyspnea. Conversely, PE can complicate acute pulmonary diseases illness (eg, emphysema, pneumonia). Thus, the presence of another diagnosis does not completely exclude the possibility of PE. Wheezing is uncommon in PE, and may suggest an exacerbation of pre-existing lung disease such as asthma or chronic obstructive pulmonary disease. However, hypoxemia or respiratory distress out of proportion to obstructive symptoms or wheezing should prompt consideration of PE.

Myocardial ischemia or infarction – Cardiac chest pain is typically not pleuritic and evidence of myocardial ischemia or infarction can be seen on electrocardiography (ECG). While troponin elevation can suggest cardiac chest pain, this can also be seen in acute PE.

Pneumothorax – While acute pleuritic chest pain and dyspnea due to pneumothorax may mimic PE, pneumothorax should be apparent on chest imaging.

Vasculitis – Unexplained dyspnea, pleuritis, and hemoptysis can be presenting symptoms of both PE and pulmonary vasculitis. The presence of an interstitial pattern on chest radiograph in a patient with an underlying rheumatologic condition (eg, scleroderma) may distinguish vasculitis from PE.

Musculoskeletal pain – Acute chest wall pain may mimic the pleuritic pain of PE. However, in the absence of a clear history of injury, musculoskeletal pain should be considered a diagnosis of exclusion when PE remains on the differential diagnosis.

In healthy individuals, the pericardial cavity contains 15 to 50 mL of an ultrafiltrate of plasma.

Diseases of the pericardium present clinically in one of several ways:

● Acute and recurrent pericarditis

● Pericardial effusion without major hemodynamic compromise

● Cardiac tamponade

● Constrictive pericarditis

● Effusive-constrictive pericarditis

Epidemiologic studies are largely lacking, and the exact incidence and prevalence of acute pericarditis are unknown. However, acute pericarditis is recorded in approximately 0.1 to 0.2 percent of hospitalized patients and 5 percent of patients admitted to the emergency department for nonischemic chest pain [1,2].

In developed countries, most cases of acute pericarditis are considered of possible or confirmed viral origin, although the exact etiology of most cases remains undetermined following a traditional diagnostic approach [2].

Patients with HIV infection treated with antiretroviral therapy who develop acute pericarditis have an etiologic spectrum very similar to non-HIV-infected patients. However, HIV infection itself, along with tuberculosis, persist as major causes of acute pericarditis in developing countries or in patients without access to antiretroviral therapy.

Acute pericarditis can present with a variety of nonspecific signs and symptoms, depending on the underlying etiology. The major clinical manifestations of acute pericarditis include [2,4]:

● Chest pain – Typically sharp and pleuritic, improved by sitting up and leaning forward. (See 'Chest pain' below.)

● Pericardial friction rub – A superficial scratchy or squeaking sound best heard with the diaphragm of the stethoscope over the left sternal border. (See 'Pericardial friction rub' below.)

● Electrocardiogram (ECG) changes – New widespread ST elevation and PR depression. (See 'Electrocardiogram' below.)

● Pericardial effusion – A pericardial effusion is a common feature of pericarditis but is not required for diagnosis. (See 'Echocardiogram' below.)

Chest pain — The vast majority of patients with acute pericarditis present with chest pain (>95 percent of cases) [5]. The chest pain of pericarditis must always be distinguished from other common and/or life-threatening causes of chest pain (table 1) such as myocardial ischemia, pulmonary embolism, aortic dissection, gastroesophageal reflux disease, and musculoskeletal pain.

Chest pain that results from acute pericarditis is typically fairly sudden in onset and occurs over the anterior chest. Unlike pain due to myocardial ischemia, chest pain due to pericarditis is most often sharp and pleuritic in nature, with exacerbation by inspiration or coughing [2]. One of the most distinctive features is the tendency for a decrease in intensity when the patient sits up and leans forward. This position (seated, leaning forward) tends to reduce pressure on the parietal pericardium, particularly with inspiration, and may also allow for splinting of the diaphragm.

Radiation of chest pain to the trapezius ridge has also been considered to be fairly specific for pericarditis. In some patients, dull, oppressive pain may occur; in such cases, it is more difficult to distinguish pericarditis from other causes of chest pain.

Chest pain is likely to be present in cases of acute pericarditis caused by infection, but may be minimal or absent in patients with uremic pericarditis or pericarditis associated with a rheumatologic disorder (although in some patients, pleuritic chest pain and pericarditis can be the initial presentation of systemic lupus erythematosus).

Pericardial friction rub — The presence of a pericardial friction rub on physical examination is highly specific for acute pericarditis (movie 1). Classically, pericardial friction rubs are triphasic, with a superficial scratchy or squeaking quality. Pericardial friction rubs are often intermittent, with an intensity that tends to wax and wane, and are best heard using the diaphragm of the stethoscope.

Pericardial friction rubs, which occur during maximal movement of the heart within its pericardial sac, are said to be generated by friction between the two inflamed layers of the pericardium. However, this commonly offered explanation for its mechanism may be an oversimplification, as patients with a pericardial effusion may also have an audible friction rub.

The classic pericardial friction rub consists of three phases, corresponding to movement of the heart during atrial systole, ventricular systole, and the rapid filling phase of early ventricular diastole. Patients in atrial fibrillation lack atrial systole, and therefore will have a two-phase rub. Additionally, for uncertain reasons, some rubs are present only during one (one component) or two phases (two components) of the cardiac cycle [6]. In a review of auscultation and phonocardiography in 100 patients with a pericardial rub, the rub was triphasic in 52 percent of patients, biphasic in 33 percent, and monophasic in 15 percent [6].

Pericardial rubs may be localized or widespread, but are usually loudest over the left sternal border [6]. The intensity of the rub frequently increases after application of firm pressure with the diaphragm, during suspended respiration, and with the patient leaning forward or resting on elbows and knees (picture 1). This last maneuver is designed to increase contact between visceral and parietal pericardium, but is seldom used in practice since it is cumbersome for the patient.

Friction rubs tend to vary in intensity and can come and go over a period of hours; therefore, the sensitivity for detection of a rub is variable and depends in large part on the frequency of auscultation. Pericardial rubs may be easier to hear in patients without a pericardial effusion, but this finding is not universal and is not well documented. In a report of 100 patients with acute pericarditis, a pericardial rub was present in 34 of 40 (85 percent) without an effusion [7]. This prevalence is considerably higher than the 35 percent incidence of friction rubs reported in another series [5].

Suspension of respiration during auscultation permits distinction of a pericardial friction rub from a pleuropericardial or pleural rub.

Electrocardiogram — The ECG in acute pericarditis may evolve through as many as four stages with highly variable temporal evolution of ECG changes.

The four typical stages of ECG changes ( figure 1) in patients with acute pericarditis include:

● Stage 1, seen in the first hours to days, is characterized by widespread ST elevation (typically concave up) with reciprocal ST depression in leads aVR and V1 (waveform 1). There is also frequently an atrial current of injury, reflected by elevation of the PR segment in lead aVR and depression of the PR segment in other limb leads and in the left chest leads, primarily V5 and V6. Thus, the PR and ST segments typically change in opposite directions. PR segment deviation, which is highly specific, though less sensitive, is frequently overlooked.

The TP segment is recommended as the baseline for comparison when measuring both PR and ST segment changes in acute pericarditis.

● Stage 2, typically seen in the first week, is characterized by normalization of the ST and PR segments.

● Stage 3 is characterized by the development of diffuse T-wave inversions, generally after the ST segments have become isoelectric. It is typically seen in the subacute phase, and its duration is not well documented and likely highly variable.

● Stage 4 is represented by normalization of the ECG. It can occur directly from stage 1 in self-limited cases or with prompt response to medical therapy.

However, pericarditis does not always result in these typical ECG changes. Many patients normalize without going past stage 1. Moreover, atypical ECG changes are seen in up to 40 percent of patients with acute pericarditis [5]. Additionally, localized ST elevation and T-wave inversion occur before ST-segment normalization in a minority of patients with acute pericarditis without myocardial involvement. These changes can simulate ECG changes seen in patients with an acute coronary syndrome.

Changes in the ECG in patients with acute pericarditis signify inflammation of the epicardium, since the parietal pericardium itself is electrically inert. However, some causes of pericarditis do not result in significant inflammation of the epicardium and, as such, may not alter the ECG. An illustration of this is uremic pericarditis, in which there is prominent fibrin deposition but little or no epicardial inflammation. As a result, the ECG often shows none of the changes associated with pericarditis [8].

In a separate report comparing patients with myopericarditis and uncomplicated acute pericarditis, cardiac arrhythmias were more commonly present in patients with myopericarditis (odds ratio 17.6, 95% CI 5.7-54.1) [3]. Thus, the presence of atrial or ventricular arrhythmias is suggestive of concomitant myocarditis or an unrelated cardiac disease.

While both acute pericarditis and acute myocardial infarction (MI) can present with chest pain and elevations in cardiac biomarkers, the ECG changes in acute pericarditis differ from those in acute ST-elevation MI (STEMI) in several ways (table 2) [11].

Morphology – The ST-segment elevation in acute pericarditis begins at the J point, which represents the junction between the end of the QRS complex (termination of depolarization) and the beginning of the ST segment (onset of ventricular repolarization), rarely exceeds 5 mm, and usually retains its normal concavity (waveform 1). Although similar patterns can occur with STEMI (where ST-segment elevation also begins at the J point), the typical finding in a STEMI patient is convex (dome-shaped) ST elevation (a pattern not characteristic of acute pericarditis) that may be more than 5 mm in height (waveform 2).

Distribution – ST-segment elevations in STEMI are characteristically limited to anatomical groupings of leads that correspond to the localized vascular area of the infarct (anteroseptal and anterior leads V1 to V4; lateral leads I, aVL, V5, and V6; inferior leads II, III, and aVF) (waveform 2). The pericardium envelops the heart, and, therefore, the ST changes are more generalized and typically present in most leads (waveform 1).

Reciprocal changes – Acute STEMI is often associated with reciprocal ST-segment changes, which are not seen with pericarditis, except in leads aVR and V1.

Concurrent ST and T-wave changes – ST-segment elevation and T-wave inversions do not generally occur simultaneously in pericarditis, while they commonly coexist in acute STEMI (waveform 2). Furthermore, the evolution of repolarization abnormalities often takes place more slowly and more asynchronously among affected leads in pericarditis than in STEMI.

PR segment – PR elevation in aVR with PR depression in other leads due to a concomitant atrial current of injury is frequently seen in acute pericarditis but rarely seen in acute STEMI.

Other – Hyperacute T waves (waveform 3), new pathologic Q waves, and QT prolongation are all rare in patients with acute pericarditis but are common in acute MI.