Development of brain metastases in non-small-cell lung cancer: high-risk features

Abstract

Aim: Brain metastases (BM) are a common site of disease progression and treatment failure in non-small-cell lung cancer (NSCLC) and can be identified in up to 30–50% of patients. Although they are common, there is no standardized screening protocol for development of BM in NSCLC. Multiple clinical variables predict increased BM occurrence, and, when present, should be used to initiate screening MRI.

Materials & methods: We performed a single center retrospective review of NSCLC patients, examining BM development and overall survival. Available clinical, radiographic and molecular data were reviewed for association with BM and overall survival. A predictive model for BM development was created for multivariate analysis.

Results: Risk factors for new BM development in NSCLC included younger age, larger primary lung tumor, Karnofsky performance score (KPS) <70, pre-existing liver or bone metastases, large cell histology and family history of cancer. Factors associated with decreased OS were larger primary lung tumor, extracranial metastases at time of diagnosis, large cell histology and poorly-differentiated carcinoma histology.

Conclusion: There are multiple high risk features for developing BM in NSCLC. Each of these factors should routinely be investigated, and presence should prompt brain MRI to allow earlier diagnosis and treatment of BM.

Plain Language Summary

non-small-cell lung cancer has a high rate of cancer spread and brain metastases (cancer spread to the brain). There is currently no standardized protocol for when to obtain imaging of the brain to inspect for metastases. Our study reviewed over 1000 patients with non-small-cell lung cancer to determine which factors predisposed to development of brain metastases, and which factors impacted survival. Risk factors for development of brain metastases included younger age, larger primary lung tumor, poor functional status, liver or bone metastases, large cell histology and family cancer history. Lung cancer patients with any of these risk factors should be screened for brain metastases. Factors associated with decreased overall survival were larger primary lung tumor, extracranial metastases at time of diagnosis, large cell histology and poorly-differentiated carcinoma histology.

Plain language summary

Article highlights.

• Brain metastases are common in non-small-cell lung cancer.

• A significant proportion of patients may have brain mets at the time of cancer diagnosis.

• Although all patients with neurological symptoms should have screening for brain mets, there are other risk factors that may indicate screening.

• Patients with larger primary lung tumors, recently progressed extracranial disease, with pre-existing bone or liver metastatic disease, or with certain histologies of lung tumor should have screening for brain mets earlier.

• Molecular marker data should be routinely collected so more individualized therapies can be.

1. Introduction

Although lung cancer treatment has advanced with targeted molecular therapies, mortality remains high. Most patients present with advanced disease, and only 19% live >5 years after diagnosis [1]. Survival depends on many variables, including genetic profile, metastases and treatment itinerary, with only a 6% 5-year OS for untreated stage I NSCLC [2].

The brain is a predominant NSCLC metastatic site and site of treatment failure, and the incidence of BM has increased alongside diagnostic imaging [3–5]. An estimated 30–50% of NSCLC patients develop BM [2–8]. The US National Cancer Registry suggests that 10% or more of NSCLC patients have BM at time of diagnosis and that median survival is approximately 30% at 1 year and 8% at 3 years [5].

Guidelines for lung cancer screening and surveillance are well-established; low-dose chest CT is recommended and shown to decrease mortality in individuals with smoking history [2]. There are no robust protocols for BM screening in NSCLC, although the NCCN 2020 Guidelines recommend “brain MRI with contrast to rule out asymptomatic brain metastases in patients with stage II, III, IV disease if aggressive combined-modality therapy is being considered.” In addition, brain MRI is optional for patients with IB NSCLC, and should be considered in “high risk” patients. After initial screening, brain MRI is not recommended for surveillance in asymptomatic patients [2].

Certain biomarkers, mutations and NSCLC histological categories appear to have an impact on BM development. Larger and main bronchus lung tumors have a higher BM rate [5,6]. Lung adenocarcinoma and large-cell tumors have a higher frequency of BM than squamous histology [5,6,9]. Lower E-cadherin expression and high serum CEA level has been correlated with BM growth and prognosis for CNS metastasis [10,11]. Patients diagnosed with EGFR-mutations have more BM but favorable treatment response [4,12]. Other relevant molecular markers for NSCLC include PDL-1, RAS pathway genes, ROS1 and ALK.

We attempted to identify high-risk features for BM in NSCLC that may lead to better screening parameters.

2. Materials & methods

2.1. Patient selection

Local Institutional Review Board (University of Illinois College of Medicine at Peoria) approval was obtained prior to study initiation (IRB 1674437-5). Due to the retrospective nature of the study and large sample size, consent from patients was waived by the review board. We performed a retrospective review of adult (≥18 years of age) NSCLC patients treated at our institution during the study period (2011–2020). Patients with lung and/or brain lesions and histological data confirming NSCLC were included. Patients with small-cell lung cancer and patients without histological data were excluded. Clinical, laboratory, radiographic and procedural details were collected. Where available, molecular marker mutation status for ROS, ALK, EGFR and PDL-1 was recorded. The primary outcome was the development of brain metastases. The secondary outcome was overall survival (OS).

Cranial imaging was performed at the discretion of the oncologist or at pertinent hospital visits. Once imaging showed BM, the neurosurgical team reviewed the case for surgical candidacy. If BM were the initial neoplastic lesion, patients were screened for other metastatic disease with either contrast-enhanced CT of the body or PET scan. Biopsy of at least one lesion was required to be included in the cohort.

All radiology reports related to cancer care were reviewed. Primary lung tumor size was measured in the largest dimension. If BM were diagnosed, imaging was reviewed by the neurosurgery team. Medical records for the 3 months preceding BM diagnosis were reviewed; if a patient had progressive extracranial disease during this period, it was recorded as such.

2.2. Statistical analysis & predictive modeling

Relevant statistics were calculated for continuous and categorical variables. The data were sorted by the initial diagnosis dates; temporal validation was used for the model and non-random splitting of the original study sample was performed. A random two thirds of the data served as development training dataset; the remaining one third of the data served as a validation dataset.

Multivariate logistic regression models were then prepared to predict the risk of BM. The final model included all predictors with p-value < 0.25. The Firth method was used for small sample bias correction. The model's ability to discriminate between presence and absence of BM was evaluated. The model performance was further assessed with a Brier score. The bootstrap method was used for the internal validation and to assess outcome prediction accuracy. Various regression models assessed the relationships between patient factors, overall survival, BM characteristics and BM treatment.

The association between overall survival and patient factors were examined by weighted Cox regression instead of Cox's regression model since for some of the factors, the proportional hazards assumption was violated. The association between the size of metastases (the largest brain metastases diameter) and patient's factors were examined using linear regression model with log-transformed metastases size. The association between the number of metastases and patient factors were examined using negative binomial model. The association between craniotomy and patients factors as well as the association between need of radiation and patient factors were examined using logistic regression models.

Patients with detection of BM >3 weeks after diagnosis of lung cancer were considered to have developed new BM after NSCLC diagnosis. BM discovered within 3 weeks of NSCLC diagnosis were considered pre-existing. With this distinction in mind, Fine and Gray and predictive models were used to examine the association between BM development and other factors, with consideration of death as a competing outcome. Statistical significance was set at p < 0.05, and subdistribution hazard ratio and 95% confidence interval were reported. Analyses were performed using SAS software version 9.4 (SAS Institute Inc., NC, USA).

3. Results

The patient selection paradigm is shown in Figure 1. Of the reviewed NSCLC patients, 229/1281 (17.87%) had BM. Descriptive statistics for all patients and subgroups within the predictive model are compiled in Table 1. The average number of lesions at BM diagnosis was 3.47 ± 4.33. Nine patients were outliers (>25 lesions) and were not included in this average. These nine patients represent 0.03% of all patients with BM, and none of these patients received surgical intervention. Patients with KPS ≥70 at time of lung cancer diagnosis had a higher number of BM (p = 0.0154), as did patients with liver metastases (RR = 1.62, p = 0.0154). The average largest diameter BM was 19 (±10) mm. .

Figure 1.

Patient selection and statistical analysis process.

Table 1.

Patient demographics for all patients combined.

Variables	No BM (n = 1052) Value (%)	BM (n = 229) Value (%)	p-value
Age at diagnosis (mean ± SD)	69.6 ± 9.8	65.9 ± 10.9	<0.001
Female sex	46.8	48.9	0.557
Non-Hispanic Caucasian	99.00	99.1	0.388
KPS ≥70 at initial Dx	74.5	71.1	<0.001
Lung tumor size (mm, mean ± SD)	34.3 ± 24.3	44.9 ± 27.9	<0.001
Lung tumor histology			<0.001
Adenocarcinoma	56.5	61.4	>0.05
Squamous	31.1	13.5	<0.001
Large cell	0.3	2.7	<0.001
Other	4.6	5.8	>0.05
Poorly differentiated carcinoma	7.2	16.6	<0.001
Bone Mets at Dx	22.3	47.9	<0.001
Liver Mets at Dx	9.2	27.0	<0.001
Other Mets at Dx	22.7	43.2	<0.001
Tobacco history			<0.001
No tobacco use	9.3	14.0	0.0984
Active smoker	33.3	42.5	<0.01
Previous smoker	57.4	43.4	<0.01
Family history cancer	62.6	63.3	0.843

Bold p-values listed in tables are <0.05 threshold.

The most common BM locations were: frontal (56.6%), parietal (36%), cerebellum (31%), temporal (28.7%), occipital (24.4%) and brainstem or deep white matter (15.9%). Approximately 48% of patients had multiple BM at diagnosis. The mean time to development of new BM after NSCLC diagnosis was 14.6 months. There were no differences in OS by anatomical location of BM (p = 0.0571).

Patients with adenocarcinoma were more likely to have multiple BM locations than those with poorly differentiated carcinoma (54.3 vs. 29.4%, p = 0.0065). Patients with poorly differentiated carcinoma were more likely to have parietal lobe BM than adenocarcinoma patients (23.5 vs. 4.7%, p = 0.0065). Patients with a KPS <70 were likely to have a higher number of BM on diagnosis than those with KPS ≥70 (p = 0.0154). Patients with pre-existing liver metastases at the time of BM diagnosis were also more likely to have multiple BM (RR 1.62, p = 0.0154).

In patients with BM, 31.8% had craniotomy for resection. The average number of BM in patients receiving craniotomy was 2.24 ± 2.8, and average diameter was 45 mm. Over half of BM patients received radiotherapy (58.5%), with an average dose of 18 Gy. Forty-nine patients (19%) received both craniotomy and radiation for BM. The average number of BM in non-surgically treated patients was 4.14 ± 4.85, and average diameter 62 mm. Approximately 29% of BM patients received no surgery or radiation, most of which had advanced disease and poor KPS.

Partial molecular marker information was available for 382 patients. Statistical analysis did not show any difference in BM development or location per marker status. Thirty-six (9.4%) patients had ROS mutations, 44 (11.5%) EGFR and 40 (10.5%) ALK. 348 patients had PDL1 status available. There was no difference in OS related to presence of molecular alterations.

After adjusting for other variables, every additional year of age at time of cancer diagnosis decreased the risk of BM by 2.6%. For each mm increase in lung tumor size, the risk of BM increased by 1.4%. With adenocarcinoma as a reference, squamous cell histology was associated with reduced BM occurrence and large cell with an increase. Predictive factors for development of BM are listed in Table 2, and overall probability of development based on these factors is shown in Figure 2.

Table 2.

Predictive factors for development of brain metastases from multivariate analysis. 

Predictive factors (pre-existing BM)	HR	95% CI		p-value
Age at diagnosis	0.957	0.935	0.979	0.0002
KPS ≥70 at initial Dx	0.434	0.267	0.704	0.0007
Liver mets at Dx	3.394	2.049	5.620	<0.0001

Predictive factors (new BM)	HR	95% CI		p-value
Lung tumor size	1.024	1.016	1.033	<0.0001
Histology of lung tumor				0.0052
Adeno	Reference
Squamous	0.339	0.146	0.787	0.0119
Large cell	7.257	1.074	49.047	0.0421
Other	0.257	0.042	1.579	0.1425
Poorly differentiated carcinoma	1.615	0.736	3.542	0.2317
Bone Mets at Dx	2.043	1.143	3.655	0.0160
Active smoker	0.342	0.156	0.752	0.0076
Previous smoker	0.387	0.196	0.764	0.0062
Family history of cancer	1.968	1.052	3.681	0.0341

Data on top for patients with pre-existing BM (within 3 weeks of diagnosis of lung cancer). Data on bottom for patients with newly developed BM (>3 weeks after diagnosis of lung cancer).

Bold p-values listed in tables are <0.05 threshold.

CI: Confidence interval; HR: Hazard ratio.

Figure 2.

Estimated probability of BM diagnosis for all patients following time of NSCLC diagnosis. Vertical line represents 3-week time point after lung cancer diagnosis.

Median survival for patients with multiple BM locations was 5.0 months from lung cancer diagnosis, which was shorter than patients with a single BM (16.6 months, p = 0.0109). Patients with pre-existing BM experienced shorter median survival compared with those who developed BM later (3.26 months (95% confidence interval (CI): 2.46–6.88) vs. 18.92 months (95% CI 14.6–27.53, respectively)). Median survival for patients without BM was 33.20 months (95% CI 28.53–37.82 months). Each of these patient groups had a significantly different OS (p < 0.0015). Variables associated with death were: lung tumor size (HR 1.005, 95% CI 1.00–1.01, p = 0.0441), large cell histology (HR 3.497, 95% CI 1.408–8.684, p = 0.007), poorly differentiated histology (HR 2.553, 95% CI 1.805–3.613, p < 0.0001), bone mets (HR 1.858, 95% CI 1.352–2.553, p = 0.0001), Liver mets (HR 2.571, 95% CI 1.668–3.965, p < 0.0001), and other mets at time of diagnosis (HR 1.961, 95% CI 1.465–2.625, p < 0.0001). KPS ≥70 was protective against death (HR 0.286, 95% CI 0.229–0.357, p < 0.0001).

Survival data for each BM treatment category are visualized in Figure 3. There was no statistical difference in OS between groups receiving neither craniotomy nor radiation and those receiving craniotomy alone, however, there was a lower OS in the craniotomy alone group when compared with the radiation alone group and the radiation and craniotomy group. (p < 0.01). The median OS for patients receiving no treatment for BM in our population was 2 months, which was significantly less than patients receiving radiation, or patients receiving radiation and craniotomy (p < 0.0001).

Figure 3.

Overall survival (OS) curves. Panel (A) survival by presence or absence of BM. Panel (B) Overall survival by BM location. Occipital and Brainstem locations not included due to lower sample size. Panel (C) Overall survival by BM location – as grouped into multiple locations versus any single location. Panel (D) Overall survival grouped by type of therapy received.

4. Discussion

We performed a retrospective review of NSCLC patients with a 17.9% rate of BM. Although this is lower than most literature, up to 30% of BM are undiagnosed during life [8,13]. Our data demonstrated similar proportions of histological categories, smoking status and other variables found in the literature, but had a heavy predominance of Caucasian ethnicity.

Screening for BM is not well-protocolized, and diagnosis of NSCLC frequently occurs at advanced stages, leading to potential delay of treatment. Previous American Thoracic Society (ATS) and European Respiratory Society (ERS) recommendations state that evaluation for BM should be based on neurological symptoms [14]. Matys et al. reviewed 1074 patients undergoing resection of newly diagnosed lung tumors. Although 82.6% of patients had no neurological symptoms, each received pre-operative contrast-enhanced head CT, which detected BM in 2.1% of patients [15]. While literature suggests that most BM provoke symptoms prior to diagnosis, studies of adenocarcinoma have demonstrated that 14–20% of BM patients are asymptomatic [16,17]. Some suggest that in surgically-treated NSCLC patients with no neurological symptoms, MRI has little use due to low BM incidence [18]. In a study of adenocarcinoma and large cell carcinoma patients, brain MRI was not advised for asymptomatic stage I-II patients but was reserved for stage III where aggressive disease therapy was planned [17]. Given not all BM are symptomatic and some symptoms are not easily discerned, there is utility in identifying other predictive variables.

The average time of BM diagnosis after NSCLC in our group was 14.6 months. Average time of diagnosis of BM in NSCLC ranges from 9 to 16 months in literature [13]. A recent review of patients with history of stereotactic radiosurgery for NSCLC BM off systemic therapy for at least 90 days showed that average time to BM development after treatment discontinuation was 16 months [19]. Without a set screening protocol, BM will continue to be diagnosed at later stages and shorten OS.

Known risk factors for BM development in our cohort included size of primary lung tumor, large cell or adenocarcinoma, young age at NSCLC diagnosis, and extracranial disease progression. Although history of smoking is ubiquitously cited as increasing risk for NSCLC, there was a lower incidence of BM within our smoking sample. The reasons for this are uncertain but may be related to the heterogeneity of tobacco exposure within smokers – the lack of detail regarding tobacco use could have introduced significant data variability.

There are mixed data regarding BM occurrence by lung tumor histology type. One systematic review showed squamous pathology as an independent risk factor for development of BM; however, patient groups were dichotomized only into squamous or non-squamous [20]. Squamous histology also has a shorter OS following surgical and radiation therapy [21,22]. Conversely, other studies show higher occurrence of BM with non-squamous histology [23–25]. Literature suggests variable BM occurrence in adenocarcinoma [5,20,26]. Despite this, several studies note better survival with adenocarcinoma histology [22,27]. Our own data showed a higher incidence of BM and significant association with mortality in large cell patients on multivariate analysis (HR 7.3). Large database studies corroborate large cell and adenocarcinoma types as a risk factor for BM [25]. While no histology subtype is exempt from BM, large cell tumors may require earlier imaging to capture and institute treatment of BM.

Cancer stage and extracranial disease progression are known BM risk factors [8,28]. In our study, 68% with newly developed BM after NSCLC diagnosis had progressive extracranial disease in the preceding 3 months. Similarly, metastases to the bone and liver were associated with BM development. Our data is consistent with prior studies reporting a higher risk of developing BM in younger patients [8]. Any demonstration of extracranial disease progression, especially in younger patients, may be an indication for brain MRI.

Our dataset was not large enough to provide risk factors predisposing to BM in specific anatomical locations. Most metastases are in the cerebral hemispheres (∼80%) and only 5–15% in the cerebellum/brainstem [13]. The frontal lobe was our most common location, which corresponds to existing literature [15]. Although there was a trend toward increased OS in patients with temporal BM location, statistical significance was not achieved (Figure 3, p = 0.0571). OS was higher in patients with a single BM (median 16.6 months) vs. multiple BM (5 months) (p = 0.0109).

We identified family history of cancer as a risk factor for BM. This may be related to individual immunologic behavior, allowing a favorable tumor micro-environment [29–31]. These responses, and hereditary factors impacting fibrogenic and angiogenic signaling likely play a role in development – though exact mechanisms linking family history and BM are not delineated.

Of the currently targeted mutations in NSCLC, there are eight driver gene-related approved therapies. The ALK gene mutation represents <10% of NSCLC but responds to tyrosine-kinase inhibitor therapy, with improvements in progression-free survival and CNS progression [8,32]. The NCCN guidelines recommend initiation of alectinib, brigatinib, or lorlatinib in ALK mutated NSCLC, with transition to another agent if the disease progresses [2]. Screening for this mutation early after diagnosis would afford the opportunity to prolong survival and reduce CNS progression [8,32,33]. In one large retrospective review, EGFR mutations in adenocarcinoma were shown to independently predict BM development [19,28]. As a result of this relationship, the authors recommended brain MRI one year after EGFR mutant cancer diagnosis [28]. A survival analysis of lung adenocarcinoma patients with BM showed alterations in the EGFR and ALK genes to have a favorable prognosis, perhaps in part due to heightened vigilance [34]. Early identification of EGFR mutant lung cancer allows for targeted medical and radiotherapy, which improves survival [4,12]. This literature makes ALK mutation identification key in any patient with NSCLC, especially with BM. Another valuable immunological marker is PD-L1. Although some earlier trials involving PD-L1 in NSCLC excluded patients with BM, approximately 25% of NSCLC patients have an elevated marker level and targeted therapy prolongs OS [35,36]. Approximately 27% of our patients had PD-L1 testing, with an average expression of 22%. Even though molecular and immunotherapy show promise, most chemotherapeutic agents and large monoclonal antibodies have difficulty crossing the blood–brain barrier and are less effective for addressing BM [12,13]. Less than 30% of our patients had molecular alteration data available, and statistical relationships for specific mutations could not be calculated.

The OS in NSCLC patients with BM is poor and, without treatment, survival is typically <3 months [37]. In general, patients with BM have better KPS and longer survival with early, aggressive treatment with surgery and radiation [7,13,38–40]. This combination of treatments is considered standard of care. A meta-analysis of 27 trials demonstrated that combined surgery and whole-brain radiotherapy improved OS [41]. Some authors have recommended prophylactic cranial irradiation for new lung cancer diagnoses, which may reduce the incidence of BM, although this is not standard for most lung cancers [42,43]. The treatment combination of craniotomy and post-op radiation showed the longest OS in our study group. Our data had a higher proportion of multiple metastases and less infratentorial disease, both variables predict decreased OS after craniotomy [40]. Higher rates of death in our patients seen with greater size of primary lung tumor, large cell histology, poorly differentiated histology and presence of extracranial metastatic disease at the time of diagnosis suggests need for appropriate counseling and early therapy to prolong survival.

There are several recent reviews that have identified some similar risk factors to our own patient population [44–47]. A meta-analysis with >10,000 NSCLC patients noted that patients with adenocarcinoma, advanced local tumor, EGFR and KRAS mutations, as well as other factors, had a higher rate of BM [44]. Although most studies suggest that certain histological patterns have higher BM risk – not all agree [5,20,23–26,45]. One systematic review remarked that machine deep learning can link common clinical and molecular risk factors for BM with MRI details [47].

Even with aggressive patient selection for treatment, earlier disease detection is paramount. In one of the few studies addressing early BM screening, Kim et al. looked at whether early brain MRI can improve quality of life and OS [17]. Recently diagnosed NSCLC patients were compared with a control NSCLC group who only underwent brain MRI if they demonstrated neurological symptoms. BM were detected in 20.8% of the study group but only 4.6% of the control group, suggesting there may be a substantial sub-set of asymptomatic patients with BM not being evaluated [17]. Despite these findings, there was no significant difference in survival time with earlier screening, though authors were statistically limited by a smaller control group [9].

Currently, there is no definition describing “early” screening for NSCLC BM. Median survival in patients with BM with treatment ranges from 6–12 months [40]. Our patient population mirrored this with a median survival of 3.26 months in patients with pre-existing BM and 18.9 months in patients who developed BM after NSCLC diagnosis. This disparity reinforces the benefit of treating BM soon after identification.

5. Conclusion

Based on our data, we propose that screening brain MRI should be performed in NSCLC patients with any extracranial disease progression (special consideration to bone and liver metastases), and in those with proven large cell histology. Multiple BM are common in NSCLC, especially in adenocarcinoma patients and those with pre-existing liver metastases. Multiple BM reduce OS and patients should be counseled accordingly during care decisions.

6. Limitations

Despite our sample size, the study has multiple limitations. Our data did not examine specific lung tumor treatment choices and their effect on BM development, representing a potentially modifiable disease factor. Although there is emerging data that specific mutations in NSCLC impact BM occurrence, our sample was not large enough to draw conclusions. Growing literature about the ALK pathway's involvement in development of BM supports the importance of screening for and using guided treatment in this NSCLC mutation. It is unfortunate that our data availability did not allow for confirmation of the significance of this finding in the BM population. These findings would have increased the current relevance of our data. The impact of family history of cancer on BM is not mechanistically understood and limits our knowledge. Finally, details regarding presence or absence of neurological symptoms at the time of brain imaging were not consistently recorded. This would have been valuable for screening recommendations in the asymptomatic population.

Author contributions

All authors contributed to the study conception and design. Material preparation and data collection were performed by N Winslow, J Boyle and W Miller. Data analysis was graciously performed by our consulting statistician, Y Wang. The first draft of the manuscript was written by N Winslow and J Boyle. All authors read and approved the final manuscript. A Tsung supervised all portions of the study and edited all the above work.

Financial disclosure

The authors have no financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Competing interests disclosure

The authors have no competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Writing disclosure

No writing assistance was utilized in the production of this manuscript.

Ethical conduct of research

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the local institutional review board prior to initiation.

Data availability statement

Data from our study are not publicly available but can be reviewed in part upon request from the corresponding author. This study does not include the original results of a clinical trial, and does not include data from secondary analysis of a clinical trial.