Between April 2019 and November 2022, 119 patients with MSCC from lung cancer were prospectively collected and enrolled at our hospital. Patients’ demographics, received treatments, tumor status, comorbidities, mental health status, and HRQoL were collected and analyzed in the study. The inclusive criteria were carefully established to ensure the selection of suitable participants for this study. First, patients were required to have a confirmed diagnosis of lung cancer through the use of tissue biopsy. Additionally, the criteria demanded that patients exhibit MSCC, which could be verified through either tissue biopsy or various radiographic imaging techniques, such as radionuclide bone scan, magnetic resonance imaging, or computed tomography. Patients experienced corresponding symptoms, such as back pain or neurological disorders, resulting from the metastatic spinal disease. Conversely, certain exclusion criteria were implemented to maintain the integrity and reliability of the study. First, patients who were 20 years of age or younger were excluded to maintain a focus on the adult population affected by this condition. Furthermore, individuals who expressed a lack of willingness to participate were ineligible to ensure a committed and engaged study group. Patients with psychiatric disorders that hindered their ability to cooperate with healthcare professionals were also excluded, as their level of collaboration is crucial for accurate data collection. Similarly, individuals who displayed unconsciousness or profound impairment of self-identity, spatial awareness, temporal perception, and expression of well-being were deemed unfit for the study. Additionally, patients with a predicted life expectancy of less than 3 months were excluded, as their ability to contribute meaningful data over the study period would be limited. Finally, patients who were lost to follow-up were not considered, as their absence from the study would result in incomplete data. The patient’s flowchart is provided in Supplementary Figure 1 . The study protocol was approved by the Ethics Committee of our hospital and informed written consent was obtained from all patients. The data were all anonymous, and the study complied with the Declaration of Helsinki.

Patients underwent a posterolateral approach for decompression of the vertebral canal and internal stabilization of the spine. The procedure involved the following steps: first, a fluoroscopy-guided incision was made along the posterior midline of the skin, followed by subcutaneous separation. The percutaneous pedicle screw fixation technique was employed to place screws through the fascia. To expose the posterior aspect of the spine, the surgeons made a small incision at the posterior center, cutting through the fascia and separating the muscles. The specific resection area depended on the extent of the tumor. After vertebrectomy, the vertebral body was reconstructed using techniques such as Kirschner wire cement, titanium mesh, or expandable interbody fusion cage. Finally, intra-fascial fixation rods were inserted and secured. The muscles and deep fascia were meticulously sutured using absorbable barbed thread, and the skin incision was closed using intradermal sutures. A case report illustrating the surgical site before and after surgery is depicted in Figure 1 .

The primary variable of interest in this study was the HRQoL, which was assessed using the Functional Assessment of Cancer Therapy—General (FACT-G) questionnaire. The FACT-G is a reliable and valid instrument commonly used to evaluate HRQoL among lung cancer patients (16). Scores on the FACT-G range from 0 to 108, with higher scores indicating better overall HRQoL. In this study, poor HRQoL was defined as a FACT-G score of less than 60. Previous research has shown that FACT-G scores tend to peak at 3 months postoperatively and remain relatively stable during the follow-up period for patients with spinal metastasis (17). Consequently, patients were asked to self-report their current well-being as part of the FACT-G score collection 3 months after discharge from the hospital.

Based on a thorough literature review and data availability, several potential risk factors were included in the analysis. These factors encompassed patients’ demographics (e.g., age, sex, marital status, religious belief, education, smoking, and drinking status, time since cancer diagnosis, and uncompleted life goals), received treatments (e.g., surgery at the primary cancer site, radiotherapy, chemotherapy, and targeted therapy), tumor status (e.g., Eastern Cooperative Oncology Group (ECOG) score, visceral metastasis, and financial burden due to cancer treatments), number of comorbidities, and mental health status (e.g., anxiety and depression scores). Time since cancer diagnosis was defined as the interval between patients’ study participation date and the date of their cancer diagnosis. The number of comorbidities represented the cumulative sum of various medical conditions, including hypertension, diabetes, coronary heart disease, chronic liver disease, chronic renal disease, and asthma. Anxiety and depression scores were assessed using the Hospital Anxiety and Depression Scale (HADS), which is a widely used tool for evaluating mental health. All patients’ demographic and clinical characteristics were collected before surgery during their hospitalization.

The least absolute shrinkage and selection operator (LASSO) regression analysis combined with 10-fold cross-validation was used to select variables for the development of the nomogram. One of the major advantages of LASSO regression is its ability to simultaneously perform variable selection and regularization (18, 19). This means that LASSO not only identifies the most relevant variables for prediction but also performs a degree of shrinkage, solving the problem of multi-collinearity in regression analysis and leading to improved model interpretability and reduced overfitting (18, 19).

The overall performance of the nomogram was evaluated using the Brier index, which measures the discrepancy between the actual binary outcomes and the predicted risk probabilities. The Brier index ranges from 0 to 1, with values closer to 0 indicating a more accurate prediction model. A Brier index greater than 0.25 is typically considered indicative of a poor prediction model. In addition, the discriminative ability of the nomogram was assessed using the concordance index (C-index) and discrimination slope. The C-index, ranging from 0.5 to 1.0, measures how well the nomogram predicts the outcome, with values above 0.7 suggesting a favorable estimation. Furthermore, probability density curves were employed to visualize the predicted probabilities between patients with negative and positive events in both the updated and previous nomograms (20–23). Calibration plots and a goodness-of-fit test were used to evaluate the calibrating ability of the nomogram, with a p-value greater than 0.05 indicating good calibration. Sensitivity, specificity, accuracy, and the kappa index were calculated to further assess the performance of both the updated and previous nomograms.

The clinical benefits of the nomograms were compared using decision curve analysis (DCA), which plots net benefits against different threshold probabilities (24). In the decision curve, a reference line called the treat-for-all scheme represents the maximum clinical costs, while a reference line named the treat-for-none scheme indicates no clinical benefit. A decision curve positioned further away from these reference lines suggests a greater clinical value of the prediction model. Additionally, the net reclassification index (NRI) and integrated discrimination improvement (IDI) were employed to compare the nomogram with and without the number of comorbidities.

To establish risk stratifications in this study, a cutoff point was determined based on an analysis of the threshold. The threshold was obtained through an area under the curve (AUC) analysis, where the predicted probability was plotted against the actual risk categories of “no” and “yes”. By plotting the predicted probabilities against the actual risk outcomes, researchers can identify the point where the model effectively distinguishes between the two categories. This point, known as the threshold, was subsequently used as the cutoff for risk stratification. Patients with a predicted probability of poor HRQoL below the threshold were classified into the low-risk group, while those above the threshold were classified into the high-risk group. The observed risk probabilities were calculated and compared between the low- and high-risk groups.

Continuous variables were presented as mean ± standard deviation (SD) or median [interquartile range (IQR)], while categorical variables were described as proportions. The Wilcoxon two-sample and Kruskal–Wallis tests were used to compare differences between subgroups. All statistical analyses were performed using the R programming language and SAS 9.4 software, with a significance level set at p < 0.05 (two-tailed).

A total of 119 lung cancer patients with MSCC were included in this study. The average age of the patients was 60.37 ± 10.49 years. The majority of patients were men (55.46%), were married (95.80%), had no religious beliefs (90.76%), and did not have a university education or above (76.48%). A large proportion of patients did not undergo surgery for their primary cancer site (76.47%), but most received radiotherapy (70.59%) and chemotherapy (68.07%). Many patients reported having unfinished wishes (76.47%). The comorbidity burden was relatively low, with only 15.12% of patients having two or more comorbidities. Demographic and clinical characteristics can be found in Table 1 . The mean FACT-G score was 56.56 ± 17.88, indicating very poor HRQoL compared to the general population. Based on the definition in our study, 56.30% of the patients were classified as having positive events. In terms of mental health, the mean anxiety score was 9.02 ± 4.70, and the mean depression score was 8.48 ± 4.88, suggesting that these patients experienced severe mental distress. Furthermore, 89.91% of the patients reported a moderate-to-severe financial burden due to cancer treatments, indicating a heavy economic burden during the COVID-19 pandemic.

Subgroup analysis of HRQoL revealed significant differences in FACT-G scores based on smoking status (p = 0.03), drinking status (p = 0.03), visceral metastasis (p = 0.049), surgery to primary cancer site (p = 0.02), chemotherapy (p = 0.01), having unfinished wishes (p = 0.01), ECOG score (p < 0.0001), tumor-targeted therapy (p = 0.02), number of comorbidities (p = 0.003), anxiety status (p < 0.0001), and depression status (p < 0.0001) ( Table 2 ). The same trend was observed in the subgroup analysis of the four FACT-G subscales. More detailed information can be found in Table 3 .

The LASSO method combined with 10-fold cross-validation identified three variables for the development of the nomogram: ECOG score, targeted therapy, and anxiety score ( Supplementary Figure 2 ). Moreover, this study found that the scores of the four FACT-G subscales were all significantly different distributed in the variable of the number of comorbidities ( Table 3 ), indicating this variable might also play an important role in affecting HRQoL among those patients. Furthermore, after a thorough review of the literature, this variable was also found to be significantly associated with HRQoL in the previous studies (25, 26). Thus, the variable was also included in the nomogram. As a result, four variables were included in the final nomogram. Notably, the coefficients between the four variables were all below 0.63 based on Spearman’s rank correlation analysis, indicating that no serious collinearity existed between the four variables.

The final nomogram was constructed using the four variables based on the logistic regression model ( Figure 2 ). The nomogram showed that a higher ECOG score, targeted therapy, a higher anxiety score, and a large number of comorbidities were associated with a high-risk probability of poor HRQoL. The variables were ranked by the SD, with the top three variables being the ECOG score, targeted therapy, and anxiety score. An example of using the nomogram to predict the probability of poor HRQoL in a given patient was also provided.

Validation of the nomogram showed that the addition of the number of comorbidities improved the overall performance of the nomogram, as indicated by a lower Brier index (0.13 vs. 0.15) and a higher C-index value (0.87 ± 0.10 vs. 0.85 ± 0.11) ( Table 4 ). A similar trend was also observed regarding discrimination slope (0.42 vs. 0.47, Figure 3 ), sensitivity (77.67% ± 17.94% vs. 80.23% ± 18.26%), and specificity (71.89% ± 16.97% vs. 76.41 ± 15.67%). The probability density curves were plotted for the previous nomogram ( Supplementary Figure 3A ) and the updated nomogram ( Supplementary Figure 3B ). In patients with positive events, the peak of the density curve was located at a high level of predicted probability of poor HRQoL, while in patients with negative events, the peak moved to a very low level of predicted probability. This indicates that the nomogram effectively differentiated patients with different HRQoL outcomes. Furthermore, the updated nomogram showed a larger gap between the density peaks of positive and negative events, indicating a stronger discriminatory ability. The calibration plot of the previous nomogram ( Figure 4A ) demonstrated good consistencies between the predicted and observed probability. The curve was generally close to the ideal reference line, indicating that the predicted probabilities aligned well with the actual probabilities of poor HRQoL. However, in the updated nomogram ( Figure 4B ), the curve was even closer to the ideal reference line, showcasing higher consistencies between the predicted and observed probabilities. This also suggests that the updated nomogram improved the accuracy of predicting HRQoL outcomes and provided a more reliable estimation of the true probabilities. The p-values of the goodness-of-fit test for both nomograms were above 0.05, suggesting significant consistencies between the predicted and observed probabilities in both models.

The changes in NRI and IDI were used to compare the accuracy between the previous nomogram and the updated nomogram. The categorical NRI was 0.113 (95% CI: 0.002–0.224, p = 0.046, Figure 5A ), and the continuous NRI was 0.404 (95% CI: 0.074–0.734, p = 0.016, Figure 5B ). The IDI was 0.035 (95% CI: 0.004–0.066, p = 0.027). These results indicated that the updated nomogram predicted HRQoL with greater accuracy than the previous nomogram. The clinical benefits of the updated nomogram were compared with those of the previous nomogram. DCA curves showed that the updated nomogram could better predict HRQoL, as it added more net benefits compared with the previous nomogram, especially for high threshold probabilities ( Figure 6 ).

Based on the nomogram, risk stratification was performed using a cutoff point of 50.00% for the predicted probability of poor HRQoL. Patients with a predicted probability of poor HRQoL of less than 50.00% were classified into the low-risk group, and those with 50.00% or above were classified into the high-risk group. In the previous nomogram, the corresponding observed risk probabilities in the low-risk and high-risk groups were 31.67% and 81.36%, respectively (p < 0.001, Table 5 ). In the updated nomogram, the observed risk probability decreased to 23.73% in the low-risk group and increased to 88.33% in the high-risk group ( Table 6 ), indicating that the updated nomogram had a better ability to recognize patients in the high- or low-risk groups. Moreover, the observed risk probabilities were more similar to the predicted risk probabilities among both risk groups based on the updated nomogram.