Appropriate use criteria for lumbar puncture and cerebrospinal fluid testing in the diagnosis of Alzheimer’s disease

2.1. Overview of the neuropathology underlying MCI and AD

At the microscopic level, the primary neuropathological features of AD include neuritic plaques and neuronal, especially synaptic, degeneration, together with other features, including cerebral amyloid angiopathy, neurofibrillary tangles, and neuropil threads [9,10]. Plaques are rounded lesions in the neuropil, primarily composed of aggregated β-amyloid (Aβ). The form ending at position 42 (Aβ42) is more prone to aggregation than shorter isoforms, of which Aβ40 is the most abundant [11,12]. Plaques exist as diffuse and neuritic plaques; the latter variant consists of a central core of amyloid fibrils surrounded by dystrophic neurites (processes filled with fibrillary tau protein), reactive astrocytes, and microglia [13]. Tangles are detected in the neuronal cytoplasm, in the form of paired helical filaments, primarily composed of hyperphosphorylated tau protein [14]. Intraneuronal accumulation of abnormally phosphorylated tau is believed to precede tangle formation in AD [15], with tau pathology spreading in a relatively consistent way from the brainstem to the transentorhinal region, hippocampal formation, and the neocortex [15,16].

The percentage of individuals who are cognitively unimpaired at death harboring plaques and tangles in their brains increases with age, particularly after the age of 65 years [17–21]. At the same time, it has long been known that the severity of neuropathological changes (density of plaques and tangles) varies considerably between patients with AD, and the severity of the pathology overlaps with the amounts of plaques and tangles found in cognitively unimpaired older adult patients [17,20,22,23]. The neuropathological features of amnestic MCI are intermediate between the neurofibrillary changes of aging and the pathologic features of early AD dementia, including the presence of mixed pathologies [24–26]. In addition to plaques and tangles, varying degrees of α-synuclein, TDP-43, cerebrovascular, and other pathologies such as hippocampal sclerosis and microinfarcts are often seen in these patients across the AD continuum [27,28]. Thus, the frequent occurrence of AD pathology in cognitively unimpaired older adult patients, as well as the presence of non-AD pathologies, complicates proper clinical stratification of patients with cognitive disturbances based on purely clinical grounds. This should be considered in clinical studies evaluating the diagnostic performance of biomarkers.

2.2. Use of CSF testing in the context of other available diagnostic tools

Recommendations for the appropriate use of CSF must be made in the context of the many other diagnostic tools in use. First, the patient’s history obtained from an informant who knows the patient well is paramount and focuses on changes in the patient’s memory, thinking, behavior, and function over time [29]. It is important to assess common, treatable causes of cognitive complaints such as sedating medications, mood disorders, and sleep disorders. A neurological examination may reveal signs of non-AD causes of dementia. Objective neurobehavioral tests are helpful in documenting the type and severity of cognitive impairment but are not by themselves diagnostic. Routine blood test including blood counts, blood chemistries, vitamin B12 level, and thyroid-stimulating hormone level are typically recommended to rule out other causes of or contributors to cognitive impairment [30]. Either a noncontrast head computed tomography scan or brain magnetic resonance imaging is usually performed to assess for cerebrovascular disease and brain atrophy.

If, after completion of this evaluation, the clinician still has significant uncertainty as to the etiology of the cognitive impairment, certain advanced diagnostic procedures can be helpful in narrowing the differential diagnosis. In some cases, fluorodeoxyglucose PET may be helpful in distinguishing between frontotemporal dementia (FTD) and AD dementia [31,32]. Amyloid PET is used to determine whether significant quantities of neuritic plaques are present in the brain, which may support the diagnosis of AD dementia or essentially rule out AD as the etiology of a cognitive complaint [33]. CSF testing is widely used throughout clinical neurology to diagnose numerous pathologies, ranging from neuroinflammatory conditions to infectious diseases. In contrast to amyloid PET, CSF studies can evaluate for many potential diagnoses, which is particularly helpful in complex and atypical cases where the differential diagnosis list can be long.

This AUC recommends the use of CSF biomarker testing for six clinical indications deemed as appropriate (see below). However, the decision to use CSF testing is also based on the clinical judgment of the provider and by the patient’s individual situation. Where diagnostic confidence of the physician is high, the use of advanced biomarker testing may not be needed.

3.1. Overview of AUC development process

The Alzheimer’s Association convened an international expert workgroup in February 2017 to begin the development of these AUC, with Avalere Health providing technical and editorial assistance to manage the development process. The workgroup participated in teleconference meetings on a biweekly basis until December 2017.

In alignment with the Institute of Medicine’s Clinical Practice Guidelines We Can Trust recommendations on group composition, the association strived to establish a multidisciplinary workgroup comprising a variety of clinicians and other professionals with relevant expertise [34]. Each member had published extensively on topics related to the key considerations around the use of LP, such as dementia research, clinical practice and ethics, and biomarker test validation and clinical utilities. The members included four neurologists, one neuroethicist, one laboratory medicine physician, and a pathology and laboratory medicine biomarker researcher (list of member names provided in Supplementary Appendix A). Five of the members were American, and two were European (Spanish and Swedish).

The workgroup developed the clinical indications shown in the Table 1 in Section 6 through a confidential and formalized process adapted from one recommended by RAND and University of California, Los Angeles [35].

3.2. Scope and key research questions

The process began with the workgroup defining the scope and parameters of the AUC and developing key research questions to guide a systematic review of the available evidence on LP and CSF using the PICOTS (population, interventions, comparisons, outcomes, timing, and settings) framework [36] (Supplementary Appendix C).

The workgroup then developed a list of 14 clinical indications that are encountered in clinical practice based on key patient groups in whom the use of LP and CSF may be considered as part of the diagnostic process.

3.3. Systematic evidence review approach and findings

In a parallel effort, the Pacific Northwest Evidence-based Practice Center at Oregon Health & Science University (OHSU) conducted the systematic review. The primary purpose of the review was to summarize and assess the strength of evidence for the safety, diagnostic accuracy, and effect on patient outcomes of LP and CSF in diagnosing AD and MCI, in cases posed in the key research questions listed in Supplementary Appendix C.

Searches for the review were conducted using Ovid MEDLINE^® without revisions (1996 to April 2017) and supplemented with review of reference lists of relevant articles and systematic reviews.

The database search resulted in 2147 potentially relevant articles. After reviewing abstracts and titles, 448 articles were selected for full-text review, of which 75 studies and 4 systematic reviews were determined to meet inclusion criteria.

Two OHSU Evidence-based Practice Center staff reviewers independently assessed the quality of each study for inclusion. The strength of overall evidence was graded as high, moderate, low, or very low using the GRADE method (based on the quality of evidence, consistency, directness, precision, and reporting bias). In conjunction, the AMSTAR (Assessing the Methodological Quality of Systematic Reviews) method was used for systematic reviews, the U.S. Preventive Services Task Force criteria were applied to randomized trials and cohort studies, and select criteria from QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) were used for diagnostic accuracy studies [37–40].

Studies captured in the review were grouped into the following categories: (1) studies in which CSF biomarkers were evaluated in comparison with a reference standard, including autopsy neuropathological evaluation or amyloid PET imaging; (2) studies of selected patients who were initially seen at a clinic without an initial diagnosis, underwent a rigorous clinical workup, and had an expert clinical diagnosis as the reference standard against which CSF results were compared; and (3) systematic reviews and meta-analyses of these and other studies, for example, case-control studies.

3.4. Limitations of the review and the inclusion of excluded studies

In evaluating the results of the findings, the workgroup considered one possible source of bias: most studies on diagnostic accuracy reviewed compared CSF biomarker accuracy to the reference standard of clinical diagnosis instead of the more objective reference standards of amyloid PET or neuropathology. Yet, it is noteworthy that newer studies (the majority of which did not meet the study inclusion criteria) to a greater extent are able to capture the value of biomarkers by using these more objective reference standards in evaluating the diagnostic utilities of CSF Aβ42, t-tau, and p-tau. It should also be pointed out that amyloid PET has shown to have false negatives in autopsy studies [41].

In light of this limitation, the workgroup included additional studies as part of the evidence base for rating the clinical indications and developing this guidance document, as shown in Supplementary Appendix E.

3.5. Discussion of the evidence

Key Question 1: What are the reported adverse effects of LP procedure for individuals undergoing evaluation for suspected AD? What are the differences in patient subgroups?

CSF can be collected safely and reliably via LP [42]. The procedure’s safety for CSF collection, which has been documented in more than 7000 patients with suspected AD in 10 studies (see Supplementary Appendix F), [43–52] is consistent with its safety record in a wide array of neurologic disorders, which include documentation of more than 30,000 patients [42,53]. Recognition of patient- and LP-related risk factors is a key to maximizing patient safety. To minimize patient risk, it is important to evaluate the patient for potential contraindications, including use of medications that could interfere with coagulation, recent seizures, some disorders of blood clotting, intracranial lesions, impaired consciousness, and papilledema [42].

Additional best practices that the workgroup recommends in ensuring LP safety include the following: (1) the use of an atraumatic narrow bore needle (associated with less risk for post-LP headache [PLPH]), (2) avoidance of repeated attempts in difficult LP cases to reduce risk for lower back pain, and (3) avoidance of collecting more than 30 mL of CSF (the threshold for PLPH risk) [42,45,50,53,54]. Finally, fear of the procedure has been shown to be an independent risk factor that can be influenced by the attitude of the clinical staff and can be decreased by providing sensitive, matter-of-fact, verbal communication about the procedure to the individual by clinicians familiar and comfortable with LP [42,45].

Key Question 2: In persons experiencing cognitive impairment, what is the diagnostic accuracy of LP and CSF in reporting CSF amyloid (Aβ42) and tau levels (t-tau, p-tau) or ratios of analytes as indicators of AD pathology presence or absence?

In the evaluation of the diagnostic accuracy for an index test, that is, the ability to discriminate between patients with and without the target condition, it is crucial to use an accurate reference standard. Accuracy measures of the test depend on the trueness of the reference standard or how well the reference standard can identify or rule out the target condition. For many conditions, an accurate reference is not available, in which case the evaluation has to depend on expert clinical diagnosis or consensus clinical diagnosis.

For AD, the accuracy of a diagnosis based on purely clinical criteria is suboptimal, with sensitivity and specificity figures around 80% and 70%, respectively, in patients followed clinically for several years at expert research centers [30]. These figures are probably substantially lower in patients in early disease stages. The reason for this is manifold, including that clinical symptoms often are vague or uncharacteristic, and overlap with those seen in other neurodegenerative disorders. In addition, the severity of plaques and tangles load overlaps with that found in cognitively unimpaired older adult patients [17,22]. Moreover, a substantial proportion of older adult patients with clinical AD has multiple pathologies, with variable severities of other proteinopathies (TDP-43 and α-synuclein), as well as cerebrovascular disease, except for plaques and tangles [27,28].

As the pathology progresses, increased numbers of neurons are affected in characteristic patterns and precede by 10–20 years or more the clinical expression of AD dementia [55,56]. These pathologic changes are reflected by decreased concentration of Aβ42 and increased p-tau CSF concentration that reflect increased amyloid plaque deposition and increased tangle density, respectively [57–60]. Increased CSF t-Tau concentration reflects the increase, intensity, and spread of neuronal injury and neurodegeneration [60–62].

For AD diagnostics, amyloid PET ligands have been developed to assess brain amyloid status [6]. Amyloid PET is an US Food and Drug Administration– and European Medicines Agency–approved technique to identify or rule out brain amyloidosis, based on validation against autopsy studies [6,63–67]. It may, therefore, be an ideal reference standard for assessing the accuracy of the CSF biomarker Aβ42, as well as the Aβ42/Aβ40 ratio. Because Aβ neuritic plaques are one of the defining pathological features of AD, assessing the accuracy of these CSF biomarkers to identify or rule out brain amyloid deposition which is evaluated by amyloid PET may also be used as a proxy for their diagnostic accuracy for detection of AD. When amyloid PET is the reference standard for detection of brain neuritic plaque burden in assessments of the diagnostic accuracy of CSF AD biomarkers, instead of clinical diagnosis alone, there is an improved diagnostic accuracy for this aspect of AD neuropathobiologic changes, with pooled values for sensitivity and specificity of 90% and 84%, respectively [68]. Studies have shown a concordance of between 85%–95% for CSF Aβ42, alone or combined with t-tau or p-tau, and amyloid PET [58,69–73]. For more detailed discussion of the characteristics of Aβ42 and the tau proteins in their respective hallmark AD neuropathologic misfolded states can be found [59–62].

Therefore, we present the diagnostic accuracy (sensitivity, specificity, and binary discrimination power) expressed as the highest receiver operating characteristic (ROC) area under the curve value for all studies comparing the CSF biomarkers with amyloid PET and autopsy diagnosis (Supplementary Appendix E). In total, 18 publications were identified by the workgroup that presented sensitivity and specificity figures for 3697 individuals, using amyloid PET as the standard for neuritic plaque burden assessment.

Four studies were based on the same consecutive memory clinic outpatient cohort but were all included in the table because different analytical techniques for measurement of the CSF biomarkers were used (Supplementary Appendix E). These studies showed a mean sensitivity of 87.6% for CSF Aβ42 to identify, and a mean specificity of 86.2% to exclude brain amyloidosis, with a ROC area under the curve value of 0.90. The corresponding figures for the CSF Aβ42/Aβ40 ratio was a sensitivity of 96.0%, a specificity of 91.3%, and a ROC area under the curve value of 0.96.

Thirteen studies were case-control, multi-center, and other cohort studies (Supplementary Appendix E), showing a mean sensitivity of 93.2% for CSF Aβ42 to identify, and a mean specificity of 84.5% to exclude brain amyloidosis, with a ROC area under the curve of 0.933. The corresponding figures for the CSFAβ42/Aβ40 ratio were a sensitivity of 96.0%, a specificity of 88.0%, and a ROC area under the curve of 0.936.

There were seven studies evaluating the ratios between t-tau or p-tau and Aβ42 in CSF (Supplementary Appendix E). These studies showed mean sensitivities of 91.1%–92.1% to identify brain amyloidosis, mean specificities of 86.3%–89.8% to exclude brain amyloidosis, and ROC area under the curve of 0.95–0.96, for the CSF t-Tau/Aβ42 and p-tau/Aβ42 ratios. Further studies are required to determine the relative merits of the ratios t-Tau/Aβ42 and p-tau/Aβ42 compared with the Aβ42/Aβ40 ratio in detection of plaque burden and for prediction of cognitive, memory, and functional decline at all stages of AD.

Potential reasons for misclassifications may include technical variability for both CSF analyses and amyloid PET assessments, especially close to the cutoffs; amyloid PET and CSF Aβ42 partly reflect different aspects of brain amyloidosis (e.g., fibrillary vs. soluble monomeric and oligomeric forms); the use of different PET ligands and protocols; and different analytical techniques for CSF measurements across studies. Nevertheless, the CSF biomarkers showed high accuracy to identify or rule out brain amyloidosis in studies using amyloid PET as the reference standard, figures being higher than those obtained when using clinical diagnosis alone as the reference standard.

Autopsy diagnosis is generally regarded as the “gold standard” for diagnosis of AD but is not as widely available as amyloid PET. Only available on death, it is not necessarily contemporaneous with the CSF sample time, thus is less often available for use as a reference standard. Diagnostic accuracy results for five studies that included 764 subjects, which compared the CSF biomarkers with neuropathology, are summarized in the table in Supplementary Appendix E, following the amyloid PET studies. The overall mean sensitivity value for either Aβ42 alone or in various combinations with either t-Tau or p-tau is 90.0%, specificity of 84.0%, and a ROC area under the curve value of 0.92. In all of these studies, the AD diagnoses were based on neuropathology, but in most cases, the determination of normal was based on clinical evaluation, thus likely reducing the accuracy for controls. This limitation is likely responsible for the somewhat lower specificity (exclusion of AD pathology) value of 84% because the brains of approximately 20%–40% of cognitively normal older adult patients have AD pathology [74–76] and therefore the specificity estimation is likely to be falsely low. Thus, these studies show, as also demonstrated using amyloid PET as the reference standard, the improved diagnostic performance of the CSF biomarkers using neuropathology as the reference standard compared with clinical diagnosis alone.

Key Question 3: In persons with little or no cognitive impairment, what is the diagnostic accuracy (sensitivity, specificity) of LP and CSF in assessing CSF amyloid (Aβ42,) and tau levels (t-tau, p-tau) or ratios of analytes as indicators of AD presence or absence?

The accuracy for detection of AD neuropathology in persons with little or no cognitive impairment is comparable with that described previously in Key Question 2, albeit in a smaller number of study subjects. Two studies evaluated the diagnostic accuracy of CSF biomarkers in cognitively unimpaired individuals for the presence of AD neuropathology using amyloid PET imaging as the standard of truth [77,78]. The studies included 158 individuals with a mean Mini–Mental State Examination score of 29, 30 of whom were amyloid PET negative, and 128 were amyloid PET positive. The sensitivity and specificity values for CSF Aβ42 or the ratios Aβ42/Aβ40 and t-Tau/Aβ42 or p-tau/Aβ42 in these two study groups fell within the range of the values, described previously in Key Question 2, found in the 18 reported studies that compared CSF AD biomarker results to amyloid PET as the measure of amyloid plaque burden or to neuropathologic evidence for AD in a mixture of cognitively normal, MCI, and AD individuals (see Supplementary Appendix E).

Key Question 4: What is the accuracy of CSF amyloid (Aβ42) and tau levels (t-Tau, p-Tau) or ratios of analytes for predicting progression from MCI to AD? Does the literature report rates of progression from MCI to AD (in individuals with MCI who are found to have amyloid and tau levels in high risk of AD ranges)?

When MCI is associated with AD pathology based on positive CSF AD biomarkers, the risk of progression to a clinical diagnosis of AD dementia is higher than when CSF Aβ42 and tau biomarkers are normal. Conversely, individuals whose CSF Aβ42 and tau biomarkers are nonpathologic have a high likelihood of remaining free from AD dementia over the time of observation, up to 10 years thus far [79–81].

The wide range of specificity values across studies that document the predictive performance of these CSF biomarkers likely occurs due to differences across studies in the longitudinal observation time period, with specificities generally improving with increased longitudinal observation time, anywhere from 1 to 10 1 years from the time of MCI diagnosis to the time limit for the study [47,75,79–81]. Other key factors such as patient age at the time of diagnosis and disease heterogeneity are likely to impact rates of progression in an individual patient [82]. Prolonged follow-up studies for at least five years, and ideally longer, are needed to provide robust estimates of the value of biomarkers in clinical practice, and for research studies such as clinical trials. There may be value in providing estimates of the risk of progression to a clinical diagnosis of dementia within three years of a clinical diagnosis of MCI [81].

Pathological CSF AD biomarker results in MCI patients have been shown to predict progression to a clinical diagnosis of AD dementia within a broad time window of 5 to 10 years. Thus, the predictive performance of CSF Aβ42 and tau biomarkers for cognitive decline in preclinical AD and MCI patients over a more narrowly defined period of time such as 2–3 years is of growing interest for potential use clinically and in research studies such as treatment trials [83–85]. Further studies will be required to establish universal cut points that can be applicable for prediction of risk for cognitive decline within a 2- to 3-year period and to test for the comparative performance characteristics for single biomarker tests versus combinations such as the ratios t-tau/Aβ42 or Aβ42/Aβ40.

Key Question 5: What are the effects of CSF testing for suspected AD on both clinical outcomes (diagnosis) and intermediate outcomes (e.g., management with medications)?

With respect to the question on the effects of CSF testing for suspected AD on patient outcomes, this is clearly an area lacking in studies because there were no study reports on the effects of CSF testing on clinical outcomes or the use of medications. Five studies [86–90] reported on change in diagnosis and confidence in diagnosis using CSF biomarkers in 1819 individuals whose age ranged from 51 to 74 years (Supplementary Appendix G). The incidence of change in diagnosis resulting from CSF testing ranged from 7% to 27%, with the majority being a change from MCI or non-AD to a diagnosis of AD. There were 8.5% [88] and 10.3% [90] changed diagnoses from AD to non-AD or MCI in the two studies that reported this information. Confidence in diagnosis or change in confidence in diagnosis was reported in two studies; one used patient vignettes, and the other involved physician responses following receipt of CSF Aβ42, t-Tau, or p-tau181 results for their patients [87,88]. In the study that used patient vignettes, the clinician participants were significantly more confident in diagnoses with AD-consistent CSF values compared with those that were borderline or normal CSF values [87]. In the other study, 32% of physicians were more confident in their diagnoses following receipt of CSF Aβ42, t-Tau, or p-tau181 results, with the greatest increase in confidence reported for the AD-diagnosed patients (from 51% to 73%). Further studies across more diverse patient populations will be required to more fully document the impact of CSF AD biomarker test results on diagnostic decision-making and to test for the effects on patient outcomes.

7.1. Preamble

The indications discussed below are general guidelines for when CSF testing might be appropriate, which may be modified by the patient’s specific clinical situation and/or the provider’s clinical judgment. For example, the age of the patient influences decision-making in several ways. Among the oldest old, questions about the ease and feasibility of performing a LP may arise. The prevalence of amyloid positivity increases with age, and multiple brain pathology is a common finding among older patients. Therefore, interpreting a positive CSF amyloid biomarker as indicating AD is not straightforward. Even the presence of both low Aβ42 and high tau or p-tau in an older adult may be ambiguous because CSF tau may increase with age; nevertheless, CSF biomarkers do retain diagnostic value among older adult patients [98].

Co-pathology and comorbid conditions are an important consideration in using and interpreting CSF biomarkers. For example, in a patient with Alzheimer’s pathology and comorbid cerebrovascular or Lewy body pathology, or factors such as medications or medical conditions that may impair cognition, deciding how much of the clinical picture may be due to AD and how much can be explained by other conditions requires clinical judgment beyond simply obtaining a CSF biomarker test.

In the appropriate use of any biomarker test, such as CSF Alzheimer’s biomarkers, the clinician should consider the pre-test probability and the likely prevalence of disease. For example, if the pre-test probability of Alzheimer’s is high, adding a CSF biomarker test with high sensitivity and specificity will not necessarily add much to the clinical judgment.

While the workgroup did not propose an algorithm to determine at what stage of the diagnostic evaluation CSF biomarkers would be appropriate for use, nor consider their use in comparison with other diagnostic tests or biomarkers, in general, they agreed that a detailed clinical and cognitive evaluation should precede the use of CSF biomarkers for each indication. In other words, there should be an initial clinical diagnosis or differential diagnosis, followed by a determination of how CSF biomarkers might contribute to the diagnosis and to clinical decision-making.

7.2. Clinical indications

Indication 1 (inappropriate): Cognitively unimpaired and within normal range functioning for age as established by objective testing; no conditions suggesting high risk and no subjective cognitive decline or expressed concern about developing Alzheimer’s disease.

This indication refers to the use of CSF biomarker testing in a patient who lacks significant risk factors for Alzheimer’s disease, and in addition, neither the patient nor an informant reports concerns about memory or cognitive changes, and the clinical evaluation has included cognitive testing that is normal.

A percentage [10–25%] of asymptomatic people may have positive CSF biomarkers for AD, and this increases with age, particularly after 60 years [99,100]. However, it is unclear whether positive CSF biomarkers can accurately predict, if or when an individual will develop MCI or AD dementia. Therefore, CSF biomarker testing of patients who are at low risk and have normal cognition is not likely to provide a useful prognostic readout and will not assist medical decision-making.

Testing of asymptomatic individuals with CSF biomarkers may be performed in a research setting, such as screening for a prevention clinical trial, but it is not recommended as part of clinical care. This recommendation for research use is also consistent with the AUC for amyloid PET.

Indication 2 (inappropriate): Cognitively unimpaired patient based on objective testing, but considered by patient, family informant, and/or clinician to be at risk for AD based on family history.

This indication refers to patients with a clinical history and cognitive assessment that are not consistent with symptomatic AD, but who are concerned about their risk for AD because of a family history of late onset AD. The rationale is similar to that discussed for Indication 1 previously. Although a positive family history may increase the lifetime risk for AD [101], the extent of increased risk for an individual is uncertain. Testing would not assist in medical decision-making.

Indication 3 (appropriate): Patients with subjective cognitive decline (cognitively unimpaired based on objective testing) who are considered to be at increased risk for AD.

SCD occurs when a patient’s cognitive abilities decline as assessed by the patient, a family member, or a physician, but the patient’s performance on objective cognitive testing remains within the normal range. Clinical methods to operationalize SCD have been proposed [95,102]. The presence of SCD increases the risk of future cognitive decline and dementia [103], and it is also related to greater levels of neocortical Aβ-amyloid burden [104] and has been associated with AD imaging features such as lower volumes of the hippocampus [105,106], and other AD signature areas [107]. Because SCD is a complex syndrome that may be caused by multiple etiologies besides AD pathology, including other neurological or medical conditions, drug use, or psychological factors, the SCD-initiative has proposed a set of specific SCD features, under the name of SCD-plus, which are associated with an increased likelihood to be an expression of the preclinical stage of AD [95,102] and are associated with future cognitive decline [108]. The following features are associated with increased risk in an individual with SCD: persistent decline in memory rather than other cognitive domains; onset in the last 5 years; age at onset >60 years; concerns (worries) associated with SCD; feeling of worse performance than others of the same age group and confirmation of cognitive decline by an informant and presence of the apolipoprotein (APOE) ε4 genotype.

For patients with SCD, the decision to perform CSF biomarker testing should be individualized, with the strongest support for testing arising when the patient, informant, and clinician all are concerned that the patient has experienced cognitive decline. In addition, CSF testing may be helpful in this group when positive AD biomarkers might influence life decisions or negative AD biomarkers would lead to further testing for alternative etiologies. Cognitive tests are not perfectly accurate, and some patients, especially highly educated individuals, may score within the normal range for the population despite having experienced significant cognitive decline.

In some studies, people with SCD have a higher risk of progressing to significant memory loss or AD, especially those who meet criteria for SCD-plus [96]. The workgroup members recognize that there is limited evidence related to CSF biomarkers in SCD. However, studies have shown an increased rate of CSF biomarkers consistent with AD pathology in SCD [109–111]. The workgroup strongly suggests that clinicians only offer testing in this population when consistent with patient goals and after a full discussion of potential biomarker test outcomes.

Indication 4 (inappropriate): Patients with subjective cognitive decline (cognitively unimpaired based on objective testing) who are not considered to be at increased risk for AD.

This indication refers to patients with subjective cognitive decline (SCD) who have been evaluated and found by a clinician to be at low risk for AD. The workgroup recognizes that studies of SCD did not always rate the clinician’s suspicion for AD. Nevertheless, there is evidence that SCD with a low index of suspicion for AD (e.g., limited or no concerns by an informant and by the clinician) is less likely to progress to cognitive decline, MCI, or AD over time [95]. Therefore, if the clinician has low suspicion for AD, the likelihood of positive AD biomarkers is low. Thus, CSF biomarker testing would be unlikely to help in clinical decision making, regardless of the results.

Indication 5 (appropriate): Mild cognitive impairment that is persistent, progressing, and unexplained.

MCI is a syndrome defined by cognitive decline, generally preserved functional abilities, and mild deficits on formal cognitive testing in one or more cognitive domains [97,112]. The workgroup acknowledges that CSF biomarkers would be most helpful in patients with MCI when symptoms are persistent and a clinical and medical workup has failed to provide a clear explanation for the MCI symptoms. Many studies have shown that MCI is associated with an increased risk of progression to dementia [113] and that different subtypes of MCI may have different risks; for example, amnestic MCI may have a higher risk of progression to AD than nonamnestic MCI [114].

The rate of AD in MCI is approximately 50%, with a 54% diagnosis rate in one large autopsy study [26]. There is strong and consistent evidence that MCI is associated with an increased risk of an AD profile of CSF biomarkers [98,111,115] and that an AD CSF biomarker profile is associated with an increased risk of progression to dementia [47,116]. Therefore, the presence of an AD biomarker profile in MCI suggests that the symptoms are caused by AD pathology, whereas normal or non-AD levels of CSF biomarkers make AD unlikely as the cause of the cognitive decline. CSF biomarkers can thus help to stratify patients with MCI to allow therapy or interventions to be initiated appropriately and to enable better prognostication. The workgroup feels that in most cases of MCI, CSF biomarkers would be appropriate, unless some other etiology was being strongly considered.

Indication 6 (appropriate): Patients with symptoms that suggest possible AD.

Possible AD has been defined in research criteria as a designation that includes atypical AD or mixed pathology (AD plus another etiology of dementia). Atypical clinical presentations such as primary progressive aphasia (PPA—particularly logopenic variant)—and posterior cortical atrophy (PCA) [29] are most often but not always associated with AD pathology. Many studies have shown that CSF biomarkers indicate an AD profile in most cases of logopenic PPA and PCA [117–119]. CSF biomarkers can therefore provide information to help to confirm or exclude AD in atypical presentations. Mixed pathology (e.g., AD plus vascular cognitive impairment or AD plus Lewy body dementia) is often found in dementia in older individuals. The question of whether AD is contributing to dementia may be uncertain after a thorough clinical evaluation that may include structural brain imaging. The presence of an AD profile of CSF biomarkers increases the likelihood that AD is indeed contributing, while normal CSF biomarkers suggest a cause other than AD.

Indication 7 (appropriate): MCI or dementia with an onset at an early age (<65).

MCI is less common in younger age groups but still carries a risk of progression to dementia. Few studies have evaluated MCI in younger individuals, but many studies have included subjects with MCI who are younger than 65 years [98]. In these studies, CSF biomarkers have comparable sensitivity and specificity for AD. Even with the smaller evidence base for younger patients, the workgroup felt that CSF biomarkers may be used to evaluate whether AD pathology underlies cognitive symptoms regardless of the age at onset.

Although the onset of dementia before age 65 is relatively uncommon, it is often associated with atypical presentations of AD and neurodegenerative disorders such as FTD. Brain pathology in younger patients is more likely to consist of a single etiology rather than mixed pathology.

Many studies of CSF biomarkers have considered patients with younger onset of dementia, and the ability of CSF biomarkers to distinguish between AD dementia and FTD has been demonstrated in clinical and autopsy series [118,120]. In patients below age 65, many other disorders besides AD may cause dementia. AD biomarkers may increase confidence in an AD diagnosis in younger patients when there is diagnostic uncertainty. The workgroup felt that if the clinical presentation was that of typical AD with younger age at onset, CSF biomarkers might not add much more confidence to the clinician’s diagnosis.

Indication 8 (appropriate): Meeting core clinical criteria for probable Alzheimer’s disease with typical age of onset.

Probable AD dementia is diagnosed when a patient with dementia has experienced the insidious onset and slow progression of cognitive decline in episodic memory, and at least one other cognitive domain and other potential etiologies have been ruled out [29]. The clinical diagnosis of probable AD dementia is associated with a neuropathological diagnosis of AD in 70%–90% of cases [26,121,122]. Neurological disorders that cause similar clinical syndromes account for most misdiagnoses and include hippocampal sclerosis, dementia with Lewy bodies, and FTD [122,123].

Given the significant rate of AD misdiagnosis, the workgroup felt that it would be appropriate to obtain CSF biomarkers in some cases of probable AD with a typical age of onset at age 65 years and older. Clinical judgment should play a major role in determining appropriateness. For example, if the clinician was less certain about the diagnosis of probable AD or considering other etiologies, CSF biomarkers may be useful. In addition, if the patient is basing major life decisions on their diagnosis (e.g., retiring or moving), confirmation of the clinical diagnosis with CSF biomarkers may ensure that the patient is making these decisions with the most complete information. These concerns can similarly influence the decision to obtain CSF biomarkers in younger onset clinically typical AD. However, if the clinician has high confidence in the diagnosis of probable AD, there are no other major diagnostic considerations, and the patient is unlikely to make major life changes based on their diagnosis, then CSF biomarkers may not be helpful.

Indication 9 (inappropriate): Symptoms of REM sleep behavior disorder.

REM sleep behavior disorder (RBD), in isolation, is a strong predictor of the eventual development of an α-synuclein-related disorder such as Parkinson’s disease, Lewy body dementia, or multiple system atrophy [124]. The connection, if any, between RBD and AD is less clear. Therefore, the workgroup thought that AD CSF biomarkers are not relevant to diagnostic decision-making in RBD.

Indication 10 (appropriate): Patients whose dominant symptom is a change in behavior (e.g., Capgras syndrome, paranoid delusions, unexplained delirium, combative symptoms, and depression) and where Alzheimer’s disease diagnosis is being considered.

In some patients, neuropsychiatric symptoms may be the earliest and most prominent symptom of dementia [125]. Especially, in cases where an older patient has no prior history of psychiatric illness, clinicians may suspect that a neurodegenerative illness, including AD, may be the etiology of these neuropsychiatric symptoms. Some patients with atypical presentations of AD may have paranoid delusions or more complex delusions such as Capgras syndrome early in the course [126]. If the neuropsychiatric symptoms do not have an obvious explanation and AD is a diagnostic consideration, the workgroup agreed that CSF biomarkers would be appropriate.

Delirium in older adult patients often has a medical explanation, such as infection, medical illness, or metabolic upset. In situations without a clear medical cause, clinicians may be concerned that the patient has underlying AD. For example, postoperative cognitive decline may be related to underlying AD pathology [127]. As part of a workup for persistent, unexplained delirium, CSF biomarkers may be considered.

Indication 11 (inappropriate): Use to determine disease severity in patients having already received a diagnosis of Alzheimer’s disease.

Clinical methods are used to stage and follow progression in patients with AD dementia. There is no evidence that CSF biomarkers can reliably stage AD dementia. Therefore, the workgroup did not think the use of CSF biomarkers for staging dementia was appropriate.

Indication 12 (inappropriate): Individuals who are APOEε4 carriers with no cognitive impairment.

The APOEε4 allele is the most common genetic risk factor for late onset AD. The workgroup discussed this as a potential indication because it is possible for people to find out, whether deliberately through a testing service or as part of a general genetic screen, whether they are APOEε4 carriers or not. Although carrying an APOEε4 allele is associated with an increased lifetime risk of developing AD dementia, it is only one of many risk factors and is not a reliable indicator of whether an individual will develop AD dementia [128,129]. Therefore, the workgroup thought that merely carrying an APOEε4 allele was not adequate justification for CSF biomarker testing. CSF biomarkers could be used in a research setting, for example, a clinical trial, in APOEε4 carriers to help to determine whether preclinical AD pathology might be present or not.

Indication 13 (inappropriate): Use of LP in lieu of genotyping for suspected autosomal dominant mutation carriers.

In patients with a family history of ADAD, genetic testing is highly accurate in determining whether an individual is an ADAD mutation carrier. CSF biomarkers may be normal in mutation carriers before amyloid deposition [130] and therefore CSF biomarkers are not reliable indicators of whether an individual is a mutation carrier. The workgroup agreed that it was inappropriate to perform CSF biomarker testing to determine ADAD mutation status.

Indication 14 (inappropriate): Autosomal dominant mutation carriers, with or without symptoms.

Early in the disease, levels of CSF biomarkers are different in ADAD as compared with late onset AD [56], which may make interpretation of CSF biomarkers in ADAD patients complex. In addition, because CSF biomarkers change many years before the onset of dementia, abnormal CSF biomarkers in an ADAD mutation carrier may not be reliable in determining whether cognitive changes are related to AD brain pathology or another cause. Therefore, the workgroup does not recommend testing CSF biomarkers in known ADAD mutation carriers for clinical purposes at this time, although use for research purposes is supported by the workgroup.