Several impairments that are commonly cited in applications for disability are challenging to assess. With these impairments in mind, SSA requested updates on the state of the science for the potential use of biomarkers to assist in the determination process. Speakers presented research and clinical insight on major depression, PTSD, schizophrenia, fibromyalgia, rheumatoid arthritis (RA), osteoarthritis (OA), and back pain. This chapter reviews their presentations and the ensuing discussions.
Madhukar Trivedi, founding director for the Center for Depression Research and Clinical Care at the University of Texas Southwestern Medical Center at Dallas, explained that the issue with biomarkers is two-fold. He said, “We need to have a better sense of the accuracy of the biomarker used,” and “what is the replicability of the biomarker, so that it can be used in scalable forms.” He presented his work on major depression, saying it is a chronic and heterogeneous disorder, and researchers are beginning to make progress on biomarkers for diagnosis and severity of disease, but “it is still a little early” to rely on biomarkers for diagnosis of depression. However, treatment-matching biomarkers are closer for widespread use, he said. The most promising biomarkers in depression are inflammatory biomarkers (e.g., C-reactive protein, IL1-beta, gene expression) and neuroimaging biomarkers (functional magnetic resonance imaging [fMRI], electroencephalogram [EEG]), but he noted that there is increasing evidence coming from metabolic markers and the gut microbiome to help with treatment metrics.
Trivedi said, “There is now very strong evidence that a proportion of patients with depression, probably between 20 and 30 percent, have some evidence of low-level inflammation that can be identified,” with consistency across several studies (Osimo et al., 2020). Trivedi also referenced a meta-analysis review on reward processing in depressed patients compared to healthy controls, which found consistent neural aberrations in imaging in the depressed patients (Keren et al., 2018). He cautioned that “there have been significant false starts in the field when examining biomarkers for depression, primarily because most of the studies were looking at biomarkers after the studies were completed.” There are also few studies that look at biomarkers to differentiate outcomes between drug and placebo interventions. A challenge with understanding biomarkers for depression is that most studies look at a single treatment, which can give some idea of prognostic indicators, but the lack of comparison
groups makes it difficult for clinicians to identify the best treatment, he said. From his perspective:
The best biomarkers for treatment matching are those that can tell you to select treatment A and avoid treatment B. This is what is really needed in order to be used in clinical practice.
One biomarker that Trivedi thought is worthwhile for treatment selection is C-reactive protein. For example, the Combining Medications to Enhance Depression Outcomes (CO-MED) trial compared outcomes for patients who were given either a selective serotonin reuptake inhibitor (SSRI) monotherapy, or bupropion plus the SSRI monotherapy. The study found that in patients where C-reactive protein is not disturbed (< 1 mg/L), the SSRI is the more effective treatment, but if C-reactive protein is equal to or more than 1 mg/L, the bupropion-SSRI combination is the more effective treatment (Jha et al., 2017). Trivedi argued that C-reactive protein is an important biomarker to consider, as without it, “we are blindly selecting patients for treatment, which is just a trial and error process.”
As another example of using an inflammatory marker for intervention, he shared a study exploring aerobic exercise as a treatment augmentation for patients with major depressive disorder who did not respond to SSRIs (Trivedi et al., 2011). In the study, the higher-dose exercise group showed a trend of higher remission rates compared to a group with a lower dose of exercise. But, upon further examination, the researchers found that only those patients with elevated levels of TNF-alpha actually had the best improvement with exercise (Trivedi et al., 2011). In other words, Trivedi said, “The anti-inflammatory effects of exercise may be effective for [treating] depression, but only if there is some elevation in inflammation, as measured in this study, at least through TNF-alpha.”
Trivedi shared a recently published EEG study he worked on with Amit Etkin, chief executive officer at Alto Neuroscience, which used a machine-learning approach to predict antidepressant response in major depression (Wu et al., 2020). Previously, EEG machine-learning studies had challenges related to volume conduction, reduced dimensionality, and signal optimization, said Trivedi (Wu et al., 2020). However, Wu et al. (2020) were able to develop a machine-learning algorithm to address these three problems. When applied to data from an imaging-coupled, placebo-controlled antidepressant study, the algorithm predicted
symptom improvement for those patients on sertraline and predicted no improvement for those on placebo (Wu et al., 2020). The algorithm was also generalizable across study sites and equipment. These findings suggest a clinical avenue for a more tailored and personalized treatment for depression, said Trivedi.
Etkin elaborated on this study by asking, when thinking about EEG as a biomarker, how do you take this information and merge it with biology to produce some type of meaningful answer that has valuable clinical implications? A large part of the answer will be machine learning (a form of artificial intelligence), he said. Specifically, he described two approaches: supervised and unsupervised. In a supervised approach, “you take data and put it through a specific algorithm to predict a known outcome and [it] tries to understand the heterogeneity in outcomes,” said Etkin. Conversely, he said, an unsupervised approach “is much more discovery oriented, and takes the same data but uses a different type of algorithm that tries to understand the heterogeneity of the biology itself, and then relates that to the outcome.” To answer his original question, Etkin shared that when you split the participants by EEG signature, the results within quartiles of response rate for sertraline when compared to the response rate for the placebo are much different than when they are all grouped together and compared to placebo (Fonzo et al., 2019; Wu et al., 2020) (see Figure 3-1).
Etkin said EEG machine learning can help identify people who would be responsive to a particular intervention. Typically, he said,
people who would come into a clinic for treatment might have very similar symptoms, but only a minority of them would respond to treatment. Therefore, that minority might drive a drug trial to show positive or negative results.
But when the group is disaggregated, the responses to a given treatment may be much larger. Additionally, Etkin said that these “various EEG signatures have no relation to depression severity at baseline.” The brain is able to make predictions that do not relate to how the disorder is characterized clinically at baseline, he elaborated.
In summary, Trivedi said that no single biomarker is likely to account for a larger population of patients with depression. Instead, it is likely that a combination of biomarkers will be needed for the best understanding. While most depressed patients do not achieve remission with the first prescribed antidepressant, “that does not mean that the second, third, or fourth treatment will not work,” he noted. Instead of the trial-and-error process used currently, using biomarkers to match patients to treatment
can be a way to help patients achieve remission much faster. Etkin suggested that “You can train a machine-learning classifier to detect a certain subtype in one cohort and apply it to another with 90 percent accuracy, consistently.” He added that replicability and robustness are important guiding factors for biomarker development and use. Trivedi said the field cannot “just depend on prognostic or predictive biomarkers,” but really needs to “look for moderators to help differentiate treatments” in order to advance. In the future, Trivedi said, this work can be used in prevention or at early onset of treatment, ideally leading to better outcomes.
Etkin elaborated on network connectivity, describing a study by Zhang et al. (2020) that characterized individuals with depression or with PTSD using an unsupervised clustering approach to examine only their EEG data to try and understand the heterogeneity between patients. He said the study found a difference in connectivity patterns between healthy participants and those with a disease, but more notably, upon closer examination, there are consistently two clusters of patients that are quite different.
The first cluster looks fairly similar to healthy participants, he explained, but the second cluster looks completely different. Zhang et al. (2020) were able to identify these two clusters in the four different cohorts that were examined, representing two different diseases—depression and PTSD. Etkin said the connectivity differences between the two clusters are heavily focused in the frontal and parietal cortex, regions in the brain important for cognitive control and attention. But all four cohorts look similar to one another, he noted.
In the PTSD sample, the study included a population of individuals entering psychotherapy-based treatment at a VA clinic. This intervention uses a biological approach by working with the person’s brain circuitry through the course of therapy. Etkin said that the study found that just like sertraline, individuals who are subtype 1 respond well to the psychotherapy intervention, but those in subtype 2 do not really respond at all. Interestingly, he added, this is independent of therapy type because the study included two different types of therapies in the cohorts and both had the same results.
In summary, Etkin said that by taking an assessment tool (e.g., EEG) used in the clinic and systematically organizing the machine learning and development of signals, it is possible to create a series of brain signatures that can accurately inform clinical decisions in different ways. But these decisions are distinct from clinical measures, as he pointed out that these findings did not relate to baseline severity of symptoms or relate to subtyping based on clinical rounds. There is something uniquely important in the biology, he said, that cuts through the clinical heterogeneity, which has been challenging in the past. He saw the near-term applicability of this approach as exciting for treatment selection and long-term prognosis.
“Mental disorders are one of the only areas in medicine that do not have the majority of conditions diagnostically confirmed by pathological measure of illness,” declared Jeffrey Lieberman, professor and chairman of psychiatry at Columbia University. The only two exceptions are Alzheimer’s disease and narcolepsy, he added. One of the first biomarkers associated with mental disorders was identified in 1967, when cerebral plaque findings in deceased patients were correlated with dementia severity and cognition (Roth et al., 1967). This study informed the field and set the stage for further growth, especially in the area of dementias, he said. “Hopefully, other mental illnesses will be able to learn more from biomarkers,” he added.
“Schizophrenia is a particularly tragic disease because it begins when people are coming into the prime of their lives,” Lieberman said. “If they
are not able to find effective treatment, they can experience deterioration in intellectual function, leading to a chronic end stage of illness where they become functionally impaired. So, the goal is to successfully identify people as early as possible and intervene to prevent this progression,” he explained. “Biomarkers would be very valuable in this process,” he said, and he described three main themes that emerge from 100 years of research on schizophrenia. First, “Schizophrenia runs in families and so is presumed to be genetic.” Second, it involves neurochemical transmission—more specifically the neurotransmitters dopamine, glutamate, and gamma aminobutyric acid (GABA) are most prominently implicated. Finally, he said
It affects brain structure, but it does not devastate the brain diffusely. The main areas implicated through a variety of research are the midbrain—where dopamine neurons are located, the frontal cortex—associated with higher mental functions, and the hippocampus, which is critical in several ways.
Dopamine dysregulation in presynaptic trafficking and release has been conclusively demonstrated in schizophrenic patients (Laruelle et al., 2003; Sulzer et al., 2000), Lieberman said. “But its utility as a biomarker is limited because the dysregulation is only seen during acute phases of the illness and does not always distinguish people with schizophrenia from healthy controls,” he noted. Additionally, using positron emission tomography (PET) scans to detect this dysregulation is expensive and cumbersome, he added. Cassidy et al. (2019) identified neuromelanin as a potential indicator because deposition of neuromelanin in the midbrain area is also seen as a measure of excessive dopamine, said Lieberman. Neuromelanin can be detected using simpler imaging, and its presence in the mid-brain as a biomarker of dopamine has been validated in postmortem brains. Sulzer et al. (2000) also found neuromelanin to be associated with the severity of schizophrenia symptoms. Lieberman noted that the study researchers suggest that neuromelanin can be useful as a diagnostic measure and potentially as a prognostic measure as well.
“The hippocampus is one of the earliest structures in the brain to be affected by schizophrenia,” said Lieberman, “but it is a very complex area.” Based on prior research, Lieberman and his colleagues have been able to focus on the CA1 subregion of the hippocampus for biomarker efforts. The model is based on glutamate because there is potential excess release of glutamate that leads to a chain reaction resulting in potential
cell atrophy. Deterioration associated with schizophrenia begins here, he explained. From a biomarker standpoint, he said there are imaging modalities that can be used to detect this deterioration, but brain changes detected by imaging often come too late in the disease progression to have strong mitigation opportunities. Another method that is promising is using spectroscopy, Lieberman said, which involves comparing GABA peaks on spectroscopic resonance imaging between clinically high-risk patients who do not have the illness yet and healthy controls. Research shows that the measure of metabolic activity of the cells in the CA1 region of the hippocampus is elevated in patients in all aspects compared to controls, but there is no difference in volume, meaning that the disease has not yet progressed to the point of structural pathology changes.
If these high-risk patients are followed, Lieberman said, “We find that the progression is not associated with the combined signal of glutamine and glutamate, but it is associated with GABA.” The correlation with symptoms is also very high with GABA. Lieberman said he believes that the profile of these measures can be validated as predictive, but what is most predictive of severe disease is the atrophy of the brain volume. Those patients with the most loss of volume have the highest rate of progression to syndromal psychosis, he added.
Biomarkers for Functional Capacity and Disability
Philip Harvey, professor of psychiatry at the University of Miami, centered his presentation on biomarkers that he said are easily detectable, readily measurable, and have strong prognostic implications. Harvey et al. (2012) looked at people who were immediately approved for disability compensation compared to those whose claims were adjudicated and subsequently either reconsidered or dropped. What the study found, he explained, is that the majority of those who are approved receive compensation immediately, but a significant number of individuals undergo a lengthy and expensive adjudication process, even though they are awarded compensation in the end (Harvey et al., 2012). Many unsuccessful initial applicants are not denied, he said, but are unable to complete the process and would likely benefit from a more objective marker. Cognitive impairment in schizophrenia is quite substantial, Harvey noted, with the average level representing an IQ score of 63. But the challenge is that many psychiatric professionals do not routinely perform cognitive assessments. For example, Keefe et al. (2006) measured correlation between performance-based cognition and functional capacity based on interviews with caregivers or people close to the patients. The study shows that subjective assessment of disability is important, but it does not correlate with objective information, he said.
“Cognition is clearly a biomarker,” he said, and because “impairment is fully developed by the time the first psychotic symptoms are identified it can be useful as an early predictive indicator of risk for disability compensation.” Harvey noted that those with considerable cognitive symptoms at their first episode of schizophrenia have only about a 14 percent chance of functional recovery 5 years later (Robinson et al., 2004), and those symptoms should be considered a first line biomarker for prediction of SSA disability status.
Beyond cognition, Harvey added that negative symptoms in schizophrenia are also important to consider and come in several forms. These include reduced emotional experience, which is highly related to social drive and functioning; blunted affect; reduced volume of speech; and reduced intonation (Ventura et al., 2015, 2019). These symptoms are often highly visible to observers, he said, and are an early marker of risk for schizophrenia. For example, studies suggest that negative symptoms at the time of a first psychotic episode can predict 8-year functional outcomes (Ventura et al., 2015, 2019), with similar predictive power as cognition. But this assessment of negative symptoms is challenging and requires clinical experience, he said, so simply asking family members may not be enough.
He shared some emerging strategies focused on “digital biomarkers” for negative symptoms in schizophrenia, and highlighted paging strategies. This approach involves asking people where they are, who they are with, what they are doing, and how they are feeling. He said that because “classic schizophrenia often includes someone who is socially isolated and sitting at home, this kind of assessment can easily detect that.” In three studies, the predominant location of people with schizophrenia was at home (Depp et al., 2019; Granholm et al., 2020; Raugh et al., 2020). Using passive measurement strategies—in this case, global positioning system (GPS) indicators of where the patient is located—Depp et al. (2019) found a high correlation between self-reported location and GPS validation. Harvey said these studies validate self-reported studies and support that the use of GPS indicators can serve as a digital biomarker of negative symptoms in schizophrenia.
Another concern with schizophrenia, especially within the context of being at home, socially isolated and inactive, is the development of early onset comorbid conditions such as metabolic syndrome, Harvey continued. Using data collected from the Suffolk County Mental Health Project, Strassnig et al. (2017b) found that patients who were overweight at the onset of their illness were about 60 percent more likely to be obese two decades later. Additionally, Strassnig et al. (2017a) found that body mass index at illness onset predicted employment significantly at the 20-year follow-up, demonstrating the cascade of becoming overweight,
then obese, and then having physical limitations that directly interfere with being able to work.
In summary, Harvey stated that cognition, weight, and activity are biological factors that can predict unemployment. These factors can also “be easily measured, present early in the illness, and are directly relevant to work outcomes and labor force participation,” he said. He suggested “using this basic approach, paired with clinical indicators, to have a broad way to address biomarkers of disability and schizophrenia.”
Daniel Clauw, director of the Chronic Pain and Fatigue Research Center at the University of Michigan, said “There is no chronic pain condition where there is a good relationship between anything peripherally measurable (e.g., radiograph or MRI) and the presence or severity of pain.” Primarily, he said
This is because pain occurs from a number of biopsychosocial factors not typically assessed or treated in clinical practice—especially central nervous system (CNS) factors that play a prominent role with disability or chronic pain.
While there are fairly good biomarkers that can measure these specific CNS factors, Clauw questioned whether these factors should be measured at all.
“Fibromyalgia is a condition that has suffered from credibility,” he explained, adding that over the past three decades researchers have learned much more about these types of centralized pain conditions. Previously, fibromyalgia was perceived “as a discrete illness characterized by focal areas of tenderness; now, it has really become the poster child for a third new mechanism of pain” called centralized or nociplastic pain, said Clauw. This mechanism also includes other pain conditions like irritable bowel syndrome, temporal mandibular joint disorder, tension headaches, and many others. “They all have prominent pain, but there is generally nothing wrong in the area of the body where the pain is felt,” he said. There is now increasing recognition that many common chronic pain conditions are overlapping, with people being subject to multiple types of pain (Maixner et al., 2016). Essentially, Clauw said, “People are developing pain in new areas and getting new diagnostic labels, but they have not developed a fundamentally new problem. Instead, the pain seems to be driven by CNS problems.”
Patient-reported outcomes can help with the assessment and diagnosis of fibromyalgia, he said. In many cases, the disability stems beyond
pain and encompasses other factors like fatigue, memory, sleep problems, mood problems, and obesity. These collective factors, rather than analyses of the brain, can more likely predict who is disabled by chronic pain. Clauw described two studies that measured whether fibromyalgia is predictive of surgery outcomes and opioid nonresponsiveness in patients. While surgical procedures seek to ameliorate pain by reducing the nociceptive input at the periphery, surgery does not address the pain that comes from the brain. The studies found that for each 1-point increase in a fibromyalgia score, patients were less responsive to both surgery and opioids.
Brummett et al. (2015) found that among patients undergoing knee or hip surgery, each reported a 1-point increase in their fibromyalgia score that was associated with an 18 percent increase in odds of failure to improve by at least 50 percent. Similarly, Janda et al. (2015) found that among women undergoing a hysterectomy, each reported a 1-point increase in their fibromyalgia score that was associated with 7 mg increase in consumption of oral morphine. Clauw said that even after patients with diagnosed fibromyalgia had surgery, the fibromyalgia score remained a powerful predictor, suggesting that the pain is coming more from the CNS than the actual surgical site.
Other techniques, including imaging, can help researchers understand why opioids are not effective for fibromyalgia or other centralized pain states. Clauw shared that using machine learning, López-Solà et al. (2017) found functional imaging (with greater than 90 percent sensitivity and specificity) could differentiate between a patient with fibromyalgia and a healthy control. These machine-learning techniques are promising in identifying subsets of pain patients who may respond differently to different therapies, but these techniques would not be as useful in adjudicating disability. He said, in conclusion, that one of the main problems in the field of chronic pain is that there are two aspects of pain. The first is what causes the pain (e.g., peripheral problem versus CNS problem). The second aspect comes from the fact that “as patients experience the pain for a longer time, they can develop downstream consequences of pain, and these functional consequences can lead to disability and functional impairments,” he said.
Over the course of his career, Clauw estimated he has seen thousands of chronic pain patients and has found it far more likely that “patients consider disability because they have received inadequate treatment for their pain, not that they have failed to respond to proper treatment.” A confluence of social, environmental, and economic factors plays a bigger role in the exacerbation of chronic pain than just biological factors, he added. “We go to great efforts to identify the people who are faking their pain,” Clauw said, “but in my experience, I can count on one hand
the people I have seen like that.” Clauw said, “fibromyalgia patients are some of the most disabled people I have seen in rheumatology,” but noted that simply putting them on permanent disability may not be the right approach.
He suggested rethinking assessing disability in people with chronic pain by learning from other groups and using care models. For example, the VA uses a care model to identify people with chronic pain in primary care and to aggressively treat them with nonpharmacological therapies. These include manual therapies (e.g., massage, acupuncture), behavioral and psychological therapies (e.g., cognitive behavioral therapy), and exercise and movement therapies (e.g., aerobic exercise, yoga).
Two speakers discussed biomarkers for different types of arthritis. Joan Bathon, chief of the Division of Rheumatology at Columbia University, discussed RA and the current state of biomarkers. Virginia Byers Kraus, professor of medicine, orthopedics, and pathology at the Duke Molecular Physiology Institute, highlighted the challenges with OA.
Building on Clauw’s categorization of pain, Bathon said that RA is more of a nociceptive pain, as it is characterized by joint inflammation, but it can also overlap with fibromyalgia pain. Current biomarkers for RA are limited, but as a general rule, she explained that laboratory biomarkers are not independently adequate for diagnosis, prognosis, or response to treatment for RA. Bathon explained that there is a genetic predisposition to RA. However, she said, the development of RA is a slowly evolving process where patients develop antibodies and an increased level of inflammatory cytokines, which eventually lead to joint pain and swelling. Whether that clinical condition progresses to disability or joint surgery depends on several factors, such as a patient’s past medical history and disease severity. Aggressive and early treatment for RA can help, but the effectiveness and accessibility of those interventions can be undermined and limited by low socioeconomic status or poor health behaviors.
Good clinical criteria for diagnosis of RA do not currently exist, Bathon said. Rather, criteria only exist for RA classification. Because RA is a slowly evolving disease and can take different paths, diagnostic criteria are difficult to develop, she said. But two antibodies are helpful for
diagnosis: anti-citrullinated peptide antibody (ACPA) and rheumatoid factor. ACPA has high specificity for RA (more than rheumatoid factor) and has predictive value for the development of RA in a patient who has a positive antibody but no symptoms (Jansen et al., 2002; Nielen et al., 2004; Schellekens et al., 2000). The detection of RA antibodies, however, can precede symptoms by up to 10 years, she added. ACPA is a good diagnostic aid and predictor of downstream disease damage (van Gaalen et al., 2005), but it is not 100 percent specific for RA, she said.
Disease Activity Biomarkers
To assess disease activity, clinicians often rely on a patient self-assessment, a clinical assessment, and the measurement of C-reactive protein. These and other factors combine to form a composite disease activity score, said Bathon. Although there is modest reproducibility in these scores between rheumatologists, there is good reproducibility in the scores if a patient sees the same clinician for serial RA assessment. The strength of assessing the disease activity score is that it can easily be done at clinics for no extra cost, and the scores often correlate with levels of treatment response. Bathon said a laboratory biomarker cannot be substituted for disease activity scores. She added that biomarkers such as C-reactive protein or interleukin-6 are inadequate on their own. There is a commercially available multibiomarker test to measure disease activity for RA called the Vectra DA (Eastman et al., 2012), but questions remain as to whether it is better than clinical disease activity scores for predicting progression of damage on radiograph, she said.
Treatment Response Biomarkers
Treatment response biomarkers include composite scores, radiographic outcomes that can measure the effect of therapy, and patient-reported outcomes, said Bathon. Many modalities used to assess damage or disability in RA do not always correlate with the patient’s self-reported pain, but instead they indicate whether or not there is continued joint damage. Understanding whether an RA patient is in remission is also a complex process, using similar composite scores and criteria, she added.
Bathon said that there is not enough data to know whether indicators like radiographically evident joint destruction or high levels of antibodies upon RA diagnosis can predict severity of RA over time. For example, Bathon said a patient with 40 swollen and tender joints may not have
more aggressive RA than someone with just two swollen joints. She added that clinicians should try to use all the available biomarkers to assess the patient’s risk of “more profound destruction” and treat accordingly.
Finally, Bathon shared that the health assessment questionnaire (HAQ) is the gold standard at FDA for evaluating disability in RA (Buchbinder et al., 1995; Wolfe et al., 2004). The HAQ predicts functional status, work disability, cost of treatment, and more, “so it is a powerful instrument and has been proven to predict RA disease progression,” she said (Buchbinder et al., 1995; Pincus et al., 1994; Wolfe et al., 2004). Biomarkers and treatment can help steer patients toward remission and good outcomes, but outcomes are dependent on several factors beyond basic health and biology, Bathon noted.
Kraus provided a brief overview of OA, which can present in the hands, hips, knees, or spine, and can be extremely disabling. All presentations of OA increase with age, and OA often affects women more than men (Murphy et al., 2008; Oliveria et al., 1995). The lifetime risk probability of symptomatic knee OA is approximately 40 percent for men and approximately 47 percent for women, and higher for those who are obese (Murphy et al., 2008).
OA is often associated with pain, fatigue, sleep disturbance, walking disability and inactivity, as well as increased morbidity of heart disease, diabetes, and hypertension (Osteoarthritis Research Society International, 2016). OA is also associated with increased mortality of approximately 55 percent (Osteoarthritis Research Society International, 2016). Early detection of OA can help prevent disability; therefore, biomarkers are critical for OA because pain and imaging changes do not provide an early signal, Kraus emphasized.
Kraus explained that OA affects the entire joint including bone, tissues, joint lining, and articular cartilage. The associated pain comes from collateral damage to other tissues caused by the breakdown of cartilage, she said. Kraus said that CTX-II, a breakdown product of Type II collagen, is the most promising biomarker to date. High baseline CTX-II is associated with a three times higher risk of knee or hip replacement, and a nine times higher risk of knee replacement in the subsequent 2 years (Bjerre-Bastos et al., 2019).
Kraus shared several methods to assess the prognosis of worsening OA. Age and gender, she said, are poorly predictive of worsening OA. However, x-ray is a useful tool, and the more severe the x-ray change, the more likely the individual is to undergo joint replacement (Neogi et al., 2009). Other methods for predicting the risk of worsening OA include a
To better understand biomarkers in OA, Kraus et al. (2011) had developed their own nomenclature in OA, which preceded the BEST glossary, called BIPEDS, and stands for Burden of disease, Investigational, Prognostic, Efficacy of intervention, Diagnostic biomarkers, and Safety. “While systemic biomarkers can be very useful and report on the whole person, they are not often used in clinical trials because trials are often focused on a single joint with standardized imaging and pain reporting,” said Kraus. The challenges for molecular markers of OA are numerous, and the pathogenesis of OA remains complex and multifactorial. Currently, Kraus said, researchers are working on understanding the magnitude of change of a biomarker and how that relates to a clinically meaningful change and outcome in a patient. “Because there are no disease-modifying drugs or treatments for OA, the situation may worsen before it improves” she concluded.
Gwendolyn Sowa, chair of the Department of Physical Medicine and Rehabilitation at the University of Pittsburgh, introduced low back pain as a multifactorial condition, similar to other conditions discussed, making it unlikely to use one single biomarker to tell the full story. She noted that the biggest component of disability for back pain comes when the patient transitions from acute to chronic pain. While the most common diagnosis in low back pain is disc degeneration, over time, many more components of the spine are affected than just the intervertebral disc (Vo et al., 2016). As a result, it can be difficult to determine the anatomical identification of the underlying syndrome, even with imaging, she said.
For many years, Sowa said that MRI was the gold standard as a biomarker, but many abnormalities on MRI are not associated with either the patient’s symptoms or level of disability. For example, Jensen et al. (1994) found that 52 percent out of 98 asymptomatic people had a disc abnormality. Additionally, lumbar imaging in patients without underlying clinical symptoms does not improve outcomes but does increase incidence of procedures for patients (Chou et al., 2009). Without evidence of improving functional outcomes, the use of this type of traditional imaging as a biomarker is questionable, she asserted. More advanced imaging modalities are in development that may be more clinically useful, but Sowa said they are not ready yet for disability determination.
“Low back pain is a multifactorial syndrome,” Sowa reiterated. Current biomarkers fall short because they often are too subjective (e.g., pain scores), not relevant to symptoms (e.g., imaging), or do not correctly
assess function, she explained. What is needed, she said, are biomarkers with increased sensitivity to changes of disease activity in real time, and specificity to patient phenotype and individual biology. There is some evidence that circulating biomarkers that are more systemic, such as CTX-II, may provide more insight and prediction of degenerative changes than the existing modalities (Sowa et al., 2009). Protein biomarkers such as neuropeptide Y; regulated on activation, normal T-cell expressed and secreted (RANTES); and CS846 have also demonstrated significant associations with the patient’s pain score (Sowa et al., 2014). While individual biomarkers may not provide sufficient information to reflect function, Sowa said that when used in combination they might be more useful than MRI. In addition to the association with pain scores, she said markers such as RANTES, which is a systemic inflammatory biomarker, have been shown to assess functions like gait speed and the Short Performance Physical Battery.
Biomarkers for Treatment
So, Sowa asked, how are biomarkers useful to guide treatment decisions? Biomarkers have been explored for the back pain population in terms of responsiveness to activity (Sowa et al., 2014). It is biologically plausible, and Sowa et al. (2014) have found circulating changes in inflammatory markers respond to activity, which is also associated with function. Regardless of possibilities, she said a one-size-fits-all approach is clearly not going to work for patients with back pain. Many therapies and exercises have been deemed as failures when they likely were not properly titrated to that particular patient and to the patient’s disease progression, she added.
Another possibility is using a biomarker to predict the likelihood of response to a specific intervention. A small study was able to identify responders and nonresponders for lumbar epidural steroid injection (Schaaf et al., 2020). The study found that those with improved pain also improved in terms of disability, and the two groups (responders versus nonresponders) showed significant differences in protein biomarkers. She also highlighted genetic biomarkers, which may be relevant to patients with back pain, as emerging evidence shows significant differences in response to treatment between those who carry a certain single nucleotide polymorphism and those who do not (Schaaf et al., 2020).
In closing, Sowa said that low back pain is a syndrome, not a single diagnosis, so determining how to phenotype patients to guide treatment will be an important next step. There is a potential to categorize patients differently based on systemic biomarkers; this would enable testing of treatments that are tailored to a patient’s individual needs and help pre-
vent or mitigate the transition from acute to chronic pain and long-term disability, she said.
Linda Brady, director of the Division of Neuroscience and Basic Behavioral Science at NIH, and Wallace moderated discussions with the speakers to better understand the translation of their research and clinical findings to be applicable to disability determination. Brady asked about the next steps for biomarkers that have been identified to show early detection in order to develop learning algorithms for predictive treatment outcomes. Etkin said that the next step is validation, with a locked FDA compatible design. While work has been done to validate the tools, there is a need for prospective trials, aligned with what FDA has for an expectation for the biomarker, he explained. Trivedi and Lieberman agreed.
Lieberman elaborated that “once you have a positive indicator of measure, the effort to identify and validate the biomarker will likely be a high-tech next step dependent on academic medical centers.” He added that once a biomarker is validated through a sophisticated method, it can be more generalizable. This process of refinement in mental health may take a long time, but “it is not that the pathology is not characterizable, it is just that the technology required has yet to be developed,” he shared.
Brady also asked how to best balance the issue of no single biomarker being sufficient to calculate the complexity of a disorder, with the existence of some very simple measures available. Harvey replied that things are “less independent than they seem.” Prediction is key, Lieberman said; if genotyping can identify how rare variances play a role in disease progression, then patients could be spared years of medication failures. In the course of illness, it could be possible to predict who becomes severely ill and who responds to treatment, he explained. Trivedi added that both of those scenarios are also present in depression. Some biomarkers could help discern subtypes, but even within subtypes more than one biomarker may be necessary to enhance accuracy of prediction and to be able to match individuals to treatment selection, he said. But that is where he sees the most value; it is still too early for biomarkers for diagnosis of depression.
Rosenbaum asked if the biomarker technologies being discussed are routinely ordered as a standard procedure, and whether insurance would cover them. Lieberman noted that for schizophrenia, biomarkers are not part of a uniform standard of care that is routinely reimbursed, though they are done in some cases. He cautioned that “psychiatry has a checkered past, and we need to be careful about making claims until there is stronger evidence of validation.” Trivedi added that reliability of measure-
ment for depression is absolutely accurate, but he suggested part of the reason measurement is not being used is because people have not imagined a new scenario where measurement is routine, and instead they fall back on a basic interview without objective measures. Etkin argued that those who control payment and reimbursement go by what is “standard in the field,” and right now most of these technologies are not. He said a more useful question is what value is generated—both clinically and economically—by doing a certain test. He questioned what objective, prespecified targets exist that can determine value for a patient. This type of benchmark for a disease or test “would be a great thing to see,” Etkin said.
Rosenbaum also asked about the practical usefulness of this emerging science in the late stage of a disease. By the time someone is so disabled they are applying for disability through SSA, is there a need for some of these more advanced methods and markers, especially when clinical diagnosis and standard procedures are clear, she asked. Clauw noted that by the time someone in pain is contemplating disability, it is very obvious economically and socially, more so than any MRI imaging can determine. Sowa replied that she could see some usefulness for going in the reverse direction—if people are looking to get off of disability, these markers could be useful. Additionally, Kraus said that “OA is increasing so dramatically that there will not be enough joint surgeons to handle the demand,” predicting that there will be people in pain who cannot get treatment efficiently due to limited staff and resources.
Thinking about the practicality of biomarkers for SSA, Brady asked which of the biomarkers discussed so far will be cost-effective and practical for use as part of a disability determination interview. Etkin reiterated the need to define value, time, and opportunity, not just cost. Once you have the equipment, he said, EEGs can be done by anyone, and cognitive tests and blood tests are also easy to conduct and cost-effective, so the main question is how good is the biomarker in addressing the value generated through that test. Another view Etkin suggested considering is the perspective of seeing psychiatric disease as a biological disease, and whether there are issues of stigma and compliance. Etkin asked if this will change how patients see themselves and their disease. Trivedi added that “people do not pay enough attention to stigma, and the impact of stigma on a patient’s willingness to get diagnosed early.” But early diagnosis could really reduce the number of people who need SSA assistance much more than current abilities allow, he said. As a clear and easy incorporation, Harvey said that neuropsychological assessments, some of which only take 30 minutes, can save adjudication in court cases, which is where large amounts of money are often spent.
Wallace asked if emphasis on various radiological biomarkers are causing overmedication or overtreatment, therefore actually increasing
disability. Bathon agreed that this can happen. She noted that she has seen this in RA where a patient has pain and after an MRI shows some small erosions, the patient is led down a path of medications and toxicities even if he or she never develops any other pain. Sowa agreed that the same occurs for the low back pain population, where patients “are horrified upon finding out they have degenerative disc disease, but the diagnosis does not give them an understanding of pain, function, or prediction of lifelong disease.” From there, she said it is difficult to convince and empower patients that they can get back to work and their activities, and that they should not just rely on a radiographic diagnosis to define their condition.
This page intentionally left blank.