Prediction of intracranial metastases in cancer patients with headache
Article first published online: 28 MAR 2002
Copyright © 2002 American Cancer Society
Volume 94, Issue 7, pages 2063–2068, 1 April 2002
How to Cite
Christiaans, M. H., Kelder, J. C., Arnoldus, E. P. J. and Tijssen, C. C. (2002), Prediction of intracranial metastases in cancer patients with headache. Cancer, 94: 2063–2068. doi: 10.1002/cncr.10379
- Issue published online: 28 MAR 2002
- Article first published online: 28 MAR 2002
- Manuscript Accepted: 20 NOV 2001
- Manuscript Revised: 30 OCT 2001
- Manuscript Received: 3 JUN 2001
- intracranial metastases;
The current study was conducted to investigate the diagnostic value of neurologic evaluation for the prediction of intracranial metastases in cancer patients with new or changed headache.
Between February 1997 and February 2000, general practitioners and specialists referred cancer patients with new or changed headache to the Department of Neurology at the study institution. All patients underwent a structured history and neurologic examination. Magnetic resonance imaging (MRI) of the brain was used as the gold standard for determining the presence of intracranial metastases. The association between baseline patient characteristics, history variables, and variables from the neurologic examination in patients with intracranial metastases was evaluated by univariate and multivariate logistic regression analyses in combination with receiver operating characteristic (ROC) curve analyses.
Sixty-eight consecutive patients were included in the current study (48 females and 20 males). The mean age of the patients was 57 years (range, 24–88 years; standard deviation ± 13.3 years). Breast carcinoma was the primary tumor in 32 patients (47.1%) and lung carcinoma was the primary tumor in 12 patients (17.6%). Intracranial metastases occurred in 22 patients (32.4%). The occurrence of intracranial metastases was predicted in the multivariate logistic regression analyses by one baseline patient characteristic variable and 2 history variables (i.e., headache duration of ≤ 10 weeks [odds ratio (OR) of 11.0; 95% confidence interval (95% CI), 1.1–108.2], emesis [OR of 4.0; 95% CI, 1.1–14.3], and pain not of tension- type [OR of 6.7; 95% CI, 1.8–25.1]). No variable from the neurologic examination was found to add information to the prediction model. When at least one of the three predictors was present, all patients with intracranial metastases could be identified with this prediction model. MRI could be omitted in 12 patients (26%) without intracranial metastases. The ROC area under curve of this model was 0.83.
Intracranial metastases were found in 32.4% of the cancer patients with headache as the presenting symptom. Although 3 significant clinical predictors were found (headache duration ≤ 10 weeks, emesis, and pain not of tension- type), few patients could be excluded from undergoing MRI because of a low specificity. Therefore, MRI of the brain was considered to be warranted in all patients in the current study. Cancer 2002;94:2063–8. © 2002 American Cancer Society.
Intracranial metastases are found in 25% of patients dying of systemic cancer.1 In > 50% of these patients, carcinoma of the breast or lung is the primary tumor.2 Approximately 80% of brain metastases are supratentorial, 15% are infratentorial, and 5% occur in the brainstem.3 The prognosis of patients with intracranial metastases depends on patient age, site of the primary tumor, the number of brain metastases, the patient's performance status, their response to steroid treatment, systemic tumor activity, and treatment modality.4 Early diagnosis therefore is important and early treatment may reverse symptoms completely, often returning the patient to a useful life, at least for some time.5–7 Headache is a common neurologic complaint in cancer patients. One prospective study, performed in a cancer hospital, found that 15.4% of cancer patients had headache. In 30% of the patients the cause was intracranial metastases. However, no headache characteristics were evaluated in this study.2 The frequency of headache in patients with brain tumors varies from 36–80% depending on the tumor type. To our knowledge, no characteristic brain tumor headache has been reported to date.8–10 At the current time, magnetic resonance imaging (MRI) generally is accepted as the best imaging technique for intracranial metastases.11 However, the diagnostic policy in patients with systemic cancer who present with new or changed headache and who appear to have a low risk for the presence of intracranial metastases remains a matter of debate. To our knowledge no studies published to date have evaluated the significance of the history and neurologic examination in the diagnosis of intracranial metastases in cancer patients. Therefore, we prospectively evaluated cancer patients with new or changed headache who were referred by general practitioners and specialists to identify the frequency of intracranial metastases and to search for significant clinical predictors of intracranial metastases.
MATERIALS AND METHODS
Between February 1997 and February 2000, we prospectively evaluated 68 consecutive cancer patients with headache. General practitioners and specialists referred cancer patients to the Neurology departments of the St. Elisabeth Hospital and Tweesteden Hospital Tilburg if the patients reported a recently developed or changed headache. Patients were excluded for 1) previous cerebral surgery, 2) brain tumor in the past, or 3) contraindications to MRI (e.g., metal prosthesis, pacemaker, etc.). Written informed consent was obtained and the medical ethical committees of the two hospitals approved the study. Within 1 week of referral, all patients underwent a standardized history-taking and neurologic examination. Patients were asked to rate the intensity of their headache on a scale of 1–10 in which 1 was equivalent to the mildest pain and 10 was equivalent to the worst pain they had ever experienced. After the history-taking headache was classified according to the International Headache Society (IHS) criteria. A Mini Mental State Examination (MMSE) was recorded. We choose MMSE scores of 24–27 points as low normal performance and those of < 24 points as abnormal.12, 13 In addition, the Karnofsky performance status (KPS) score was recorded. Cutoff points for age, pain score, onset headache, and first visit as well as the interval between the development of the primary tumor and the first visit were designated after reviewing the results and we choose the cutoff point with the highest sensitivity and specificity. The gold standard in all patients was MRI of the brain within 1 week after the neurologic examination. In the St. Elisabeth Hospital, images were obtained with a Gyroscan T5 (Philips Medical Systems, Shelton, CT) with a superconductive magnet and a field strength of 0.5 tesla (T). In the Maria Hospital, images were obtained using a Magneton P8 (Siemens, Munich, Germany) with a permanent magnet and field strength of 0.2 T. The following scan protocol was performed: 1) sagittal T1-weighted sequences, 2) transverse T2-weighted sequences, and 3) transverse T1-weighted sequences with gadolinium diethylenetriamine pentaacetic acid (Gd-DTPA). Two experienced neuroradiologists assessed all MRI studies. The assessment focused on the presence or absence of parenchymal metastases, leptomeningeal metastases, calvaria or base of skull metastases or a combination. After diagnosis, patients with intracranial metastases received treatment according to internationally accepted neurooncology standards. Patients usually received at least one of the following treatment modalities: analgesics, steroids, and radiotherapy or surgery. Patients with intracranial metastases were seen for follow-up every 2 months until February 2000 or until their death. Patients without metastases were seen after 1 year for follow-up.
To determine predictors for intracranial metastases, univariate analysis was performed using baseline patient characteristics, standardized neurologic history variables, and neurologic examination variables. Statistical significance was assessed using the two-sided Fisher exact test. We considered a P value < 0.05 to be statistically significant. All continuous variables were dichotomized.
To assess the independence of the significant predictors for intracranial metastases, multivariate analysis was performed using multiple logistic regression analysis.
To construct a prediction model, backward stepwise logistic regression analysis was performed on the significant variables with a sensitivity > 50%. Referral pattern was not used in the multivariate regression analyses to avoid multicollinearity.
The ability to discriminate between patients with and patients without metastases was estimated using the area under the receiver operating characteristic (ROC) curve. A P value > 0.7 was interpreted as reasonable and a P value > 0.8 was considered good.
We calculated a risk estimation for each patient using the logistic regression function (y = c + b1x1 + b2x2 + … bjxj) to determine possible clinical decision-making rules for whether to perform MRI in cancer patients presenting with new or changed headache.
The statistical software programme SPSS, Version 9.0x (SPSS, Inc., Chicago, IL), was used for statistical analyses.
A total of 68 consecutive cancer patients with headache were included in the current diagnostic study. The baseline patient characteristics, history variables, and variables from the neurologic examination are shown in Table 1. The mean age of the patient cohort at the time of referral was 57 years (range, 24–88 years). Forty-eight patients were female (70.6%). General practitioners referred 25 patients (37%) and specialists referred 43 patients (63%).
|Predictor||Metastases absent (n = 46)||Metastases present (n = 22)||OR (95% CI)||P value|
|Baseline patient characteristics|
|24–58 vs. 59–88||22||(48)||15||(68)||2.3 (0.8–6.8)||0.13|
|Female vs. male||32||(70)||16||(73)||1.2 (0.4–3.6)||1.00|
|Spec vs. gen pract||22||(48)||21||(96)||22.9 (2.8–184.8)||0.0001a|
|Lung vs. breast/other||6||(13)||6||(27)||2.5 (0.7–8.9)||0.18|
|Breast vs. lung/other||24||(52)||8||(36)||0.5 (0.2–1.5)||0.30|
|Other vs. lung/breast||16||(35)||8||(36)||1.1 (0.4–3.1)||1.00|
|Interval b/w onset headache-first visit (wks)|
|0–10 vs. 11–36||30||(65)||21||(96)||11.2 (1.4–91.1)||0.007|
|Interval b/w primary tumor-first visit (mos)|
|0–62 vs. 63–244||31||(67)||19||(86)||3.1 (0.8–12.0)||0.14|
|Blurred vision||17||(37)||6||(27)||0.6 (0.2–1.9)||0.59|
|Speech disturbance||5||(11)||5||(23)||2.4 (0.6–9.4)||0.27|
|Arm/leg weakness||5||(11)||3||(14)||1.3 (0.3–6.0)||0.71|
|Gait disturbance||8||(17)||7||(32)||2.2 (0.7–7.0)||0.22|
|Pain worse at night||24||(52)||10||(46)||0.8 (0.3–2.1)||0.79|
|Pain triggered by exercise||16/44||(36)||8/21||(38)||1.1 (0.4–3.2)||1.00|
|Pain worse with Valsalva maneuver||15||(33)||9||(41)||1.4 (0.5–4.1)||0.59|
|6–10 vs. 1–5||33/44||(75)||16||(73)||0.9 (0.3–2.8)||1.00|
|Other vs. tension-type||8||(17)||12/21||(55)||5.7 (1.8–17.7)||0.004|
|Yes vs. no||38||(83)||17/22||(72)||0.7 (0.2–2.5)||0.74|
|0–60 vs. 70–100||6/43||(14)||7/22||(32)||2,9 (0.8–10.0)||0.11|
|1–23 vs. 24–30||1/45||(2)||4/20||(20)||11.0 (1.1–105.9)||0,03|
|Cranial nerve deficit||6||(13)||7||(32)||3.1 (0.9–10.8)||0.10|
|Motor disturbance||2||(4)||3||(14)||3.5 (0.5–22.5)||0.32|
|Sensory disturbance||3||(7)||2||(9)||1.4 (0.2–9.3)||0.66|
|Coordination disturbance||9||(20)||10||(46)||3.4 (1.1–4.3)||0.04|
|Babinski sign||2||(4)||5||(23)||6.5 (1.1–36.6)||0.03|
The majority of patients had a primary tumor of the breast (47.1%) or lung (17.6%), followed by colorectal carcinoma and non-Hodgkin lymphoma. The median interval between the time of diagnosis of the primary tumor and referral was 39 months (range, 0–244 months). The median interval between headache and neurologic consultation was 5.5 weeks (range, 0–36 weeks). Intracranial metastases were found in 22 patients (32.4%) based on MRI scans of the brain with gadolinium. Multiple parenchymatous metastases occurred in nine patients, leptomeningeal metastases occurred in four patients, combined leptomeningeal and parenchymatous metastases occurred in two patients, single metastases occurred in five patients (two infratentorial and three supratentorial metastases), and two patients had metastases of the base of the skull and calvaria (one patient had dural invasion). The neurologic examination was normal in four of these patients. Specialists referred 21 of the 22 patients (96%) with metastases.
Patients with metastases had at least a 10% higher score than patients without metastases for 5 of the 8 baseline patient characteristics (i.e., age, referral by a specialist, lung carcinoma, headache duration of ≤ 10 weeks at the time of neurologic consultation, and interval of ≤ 62 months between the diagnosis of the primary tumor and the first visit). In patients without metastases, there was a 10% higher score for 1 of the 8 baseline patient characteristics (breast carcinoma) compared with the patients with metastases. Two baseline patient characteristics (other primary tumor and gender) demonstrated no apparent difference of > 10% between the 2 patient groups. Patients with metastases had at least a 10% higher score for 7 of the 15 medical history variables (i.e., nausea, emesis, diplopia, speech disturbance, gait disturbance, pain not of tension type, and a KPS ≤ 60). Patients without metastases scored at least 10% higher for 2 of the 15 medical history variables (i.e., blurred vision and pain medication). For 6 of the 15 medical history variables, there was no difference > 10% (arm/leg weakness, seizure, pain worse at night, pain triggered by exercise, pain worse with the Valsalva maneuver, and pain score).
In the patients with metastases there was at least a 10% higher score for 7 of the 9 variables from the neurological examination (MMSE score of ≤ 23, apathy, papilledema, cranial nerve deficit, motor disturbance, coordination disturbance, and Babinski sign). For 2 of the 9 variables from the neurologic examination there was no difference that was > 10% (speech disturbance and sensory disturbance).
After univariate analysis of baseline patient characteristics and variables from history and neurologic examination, an association was found between intracranial metastases and 7 variables: interval between headache onset and neurologic consultation of ≤ 10 weeks (odds ratio [OR] of 11.2; 95% confidence interval [95% CI], 1.4–91.1), emesis (OR of 4.93; 95% CI, 1.6–15), pain not of tension type (OR of 5.7; 95% CI, 1.8–17.7), MMSE score of ≤ 23 (OR of 11.0; 95% CI, 1.1–105.9), apathy (OR of 10.0; 95% CI, 1.0–95.7), coordination disturbance (OR of 3.43; 95% CI, 1.1–4.3), and Babinski sign (OR of 6.47; 95% CI, 1.1–36.6).
Table 2 shows the multivariate prediction model for intracranial metastases. Three variables were found to be significant independent predictors: headache duration of ≤ 10 weeks (OR of 11.0; 95% CI, 1.1–108.2), pain not of tension type (OR of 6.7; 95% CI, 1.8–25.1), and emesis (OR of 4.0; 95% CI, 1.1–14.3). The ROC curve of this model is shown in Figure 1. The area under the curve was 0.83.
|Predictor||OR (95% CI)||P value|
|Headache ≤ 10 wks||11.0 (1.1–108.2)||0.04|
|Pain not of tension-type||6.7 (1.8–25.1)||0.004|
|ROC area||0.83 (0.72–0.93)|
Table 3 shows the calculated risk estimation of having intracranial metastases using the logistic regression function of the three predictors. When at least one of the three predictors was present, all patients with intracranial metastases could be identified. In this manner, 12 MRI scans (26%) could be omitted in patients without intracranial metastases.
|Predictor||Calculated risk score (y = 3.9 + 2.4 X1 + 1.9X2 + 1.4X3)||Metastases absent (n = 46)||Metastases present (n = 22)|
|E + P||−0.6||1||(2)||0||(0)|
|E + H||−1.0||7||(15)||5||(23)|
|P + H||0.4||3||(7)||6||(27)|
|E + P + H||1.8||1||(2)||6||(27)|
As a single predictor, emesis predicted 1 of the 22 cases of metastases (5%) and there were no negative MRI findings. As a single predictor, a headache duration of ≤ 10 weeks predicted 4 of the 22 positive MRI scans (18%) (with metastases) and 19 of 46 negative MRI scans (41%).
The combined presence of the predictors of emesis and headache of a duration ≤ 10 weeks predicted 5 of the 22 positive MRI scans (23%) and 7 of 46 negative MRI scans (15%). The combined presence of pain not of tension type and a headache duration of ≤ 10 weeks predicted 6 of the 22 cases of metastases (27%) and 3 of 46 negative MRI scans (7%). The combined presence of emesis and pain not of tension type predicted none of the 22 positive MRI scans and 1 of 46 negative MRI scans (2%).
All three predictors present predicted 6 of the 22 metastases (27%) and only 1 of 46 negative MRI scans (2%).
The current study was undertaken to investigate the incidence of intracranial metastases prospectively and to assess the diagnostic value of neurologic evaluation. Headache is a common symptom in cancer patients and access to MRI remains limited. Multivariate logistic regression analysis was used to derive significant and independent clinical predictors from the baseline patient characteristics, medical history and neurologic examination. Clouston et al.2 prospectively evaluated neurologic complications in cancer patients and found an incidence of 30% of intracranial metastases in patients with headache who were referred to the neurology department. This study was performed in a cancer hospital. General practitioners and specialists referred our patients and we found an incidence of 32.4% for intracranial metastases. This surprisingly equal incidence most likely can be explained by the fact that patients with intracranial metastases in the current study were referred mainly by specialists (21 of 22; 96%). After excluding the three most statistically significant predictors from the univariate analyses, the group of patients who were referred by specialists still had more symptoms and signs compared with patients referred by general practitioners. Moreover, we found retrospectively that 68% of the cancer patients with intracranial metastases did not have stable disease. The incidence was 20% in those patients without intracranial metastases (OR of 8.8; 95% CI, 2.8–28.0) (P < 0.0001). We speculate that those patients referred by specialists had more advanced disease and were at a greater risk of developing intracranial metastases.
The main clinical predictors in the current study were headache duration of ≤ 10 weeks, emesis, and pain not of tension type. When at least one of the three predictors was present, all patients with intracranial metastases could be identified. We have to bare in mind that by chance, with 30 variables we would expect 1 or 2 to be statistically significant at the P < 0.05 level. According to this model, MRI could be omitted in 12 patients without intracranial metastases (26%). The results of the current study demonstrate that the discriminative ability of the medical history, including headache features, was low. The neurologic examination did not appear to contribute to the prediction model.
To our knowledge there is a paucity of literature regarding headache in patients with systemic cancer. Also, to our knowledge there are no studies published to date concerning the predictive value of the medical history and neurologic examination for intracranial metastases. As in other, albeit retrospective, studies regarding headache in patients with brain tumors (both primary and metastatic),9, 10 we found headache intensities of mild to moderate severity. Emesis was present in 55% of the patients in the current study with intracranial metastases. Forsyth and Posner9 found an incidence of nausea and emesis in 48% of 111 patients (66% with metastatic brain tumors) and Suwanwela et al.10 found an incidence of 36%; both studies were retrospective in nature.
Forsyth and Posner9 described a tension-type headache in 77% of their patients and Suwanwela et al.10 described a dull/steady headache in 74% of their patients. In contrary to these studies, we found a tension-type headache in only 45% of patients with intracranial metastases in the current study. Pain not of tension-type even was found to discriminate between those patients with intracranial metastases and those without. Nocturnal headache was reported to occur in 32% of patients in the study by Forsyth and Posner9 and 71% in the study by Suwanwela et al.,10 and in the current study it was found to occur in 46% of patients.
Although the results of the current study reported three clinical predictors of intracranial metastases, we believe few patients can be excluded from undergoing MRI because of a low specificity. These findings need to be confirmed in a larger study before those identified prognostic factors may be used to identify cancer patients with headache who are at risk for intracranial metastases. Furthermore, we recommend the prospective evaluation of the stage of disease as an independent risk factor for intracranial metastases. At the current time, MRI of the brain still is warranted in all cancer patients with a new or changed headache of recent onset.
- 1Intracranial metastases from systemic cancer. Adv Neurol. 1978; 19: 575–592., .