Reply to the Comments on “Development and Validation of Hospital Mental Health Screen to Detect Psychiatric Morbidity in Medically Ill Patients in India”

Dear Commentators,

The methodology in our article, ¹ used Principal Component Analysis (PCA) as an initial step to identify the underlying dimensions of mental health in general hospital patients with medical illnesses. This helped reduce the complexity of the data, which is a widely acceptable approach ² to data reduction by sequentially using PCA and then factor analysis techniques.

Exploratory Factor Analysis (EFA) is carried out to identify and validate the latent constructs. It is a recommended approach to comprehend the relationship between theoretical constructs and the observed variables. ³ This helped us ensure that we aligned with the theoretical framework of our study and had an apt interpretation of latent constructs. Although studies highlight the importance of PCA in providing a concise and clear summary of item scores, the limitation includes a lack of using Principal Factor Analysis (PFA). ⁴ The combined approach, though rigorous, does not consider alternative approaches, such as parallel analysis in the article, which could have provided a conservative approach to factor retention. CFA is pertinent in validating the factor structure. ⁵ We used EFA to understand the latent constructs and have suggested CFA usage as a future research domain, as studies have depicted the importance of this methodological sequence in validating the factor structure. ⁶

We further add that a scree plot and eigenvalue (≥ 1) were used to retain the factors. Eigenvalue threshold (≥ 1) has been used extensively. ⁷ Scree plot facilitates the identification of the point at which there is minimal contribution of additional factors in the variance. We agree that scree plots, at times, can be ambiguous in distinguishing factors and noise. For instance, in the current study, we could have also considered a seven-factor structure if the eigenvalue threshold criteria were met for the same. There is strong evidence of the potential benefits of using scree plots with eigenvalue thresholds. This method also provides visual ease in interpretation. ⁸

The distribution of symptoms, for instance, depression symptoms, across multiple factors (F1, F3, and F6) reflects the non-homogeneous nature of symptom grouping, reflective of the complex dimensional nature of mental health constructs during naturalistic screening across medically ill patients, while excluding groups where mental health assessments might differ significantly, such as those needing palliative care, end of life care, etc., ⁹ from heterogenous groups as shown in Table 1. This approach thus provides a balanced comprehension of symptomatology. In consultation-liaison psychiatry (CLP), distinguishing between medical and psychiatric origins of symptoms can be challenging. This uncertainty often leads to a substitutive approach, where clinicians replace certain symptoms with others to address both possibilities. For example, items like Item-5 “I have trouble sleeping” or Item-4 “I am unable to concentrate on my work” is substituted with alternative symptoms to better capture the patient's condition. Indeed, the limitation of not using screening tools such as PHQ-9 or GAD-7 is acknowledged at the cost of the limited ability of these tools to screen depression and anxiety. One more limitation is not running a sensitivity analysis, which could examine how results vary with specific subgroups, such as by age, gender, or type of medical condition. We have also not assessed the severity, duration, and disability related to the medical conditions. The PCA is predominantly viewed under the lens of initial exploration with its potential role in the development of further diagnostic tools. ¹⁰

Table 1.

Distribution of Medically Ill Patients in the Study Sample.

Department	No of Patients (%)
Medicine	206 (51.9)
Surgery	146 (36.8)
Ophthalmology	20 (5)
Pulmonary medicine	12 (3)
ENT	10 (2.5)
Orthopedics	3 (0.8)

Another limitation is that the concurrent validity is not detailed amidst the lack of comprehensive screening tools in hospital settings. This was established by comparing the results from the newly developed tool with psychiatric diagnoses made by two independent psychiatrists based on clinical interviews and mental status examinations, as per ICD-10 guidelines. A Receiver Operating Characteristic (ROC) analysis was conducted, using the psychiatrists’ diagnoses as the gold standard. The total number of patients screened by the psychiatrists was 397, while 175 had psychiatric morbidity for this study.