Usually major psychiatric pathology (what in older classification systems would fall under psychosis, as in a severe mental illness with poor reality testing) is characterized by poor or impaired insight.
As a result, patients with such pathology routinely refuse help as they tend to believe that either they don’t have a problem (patients with schizophrenia or mania) or are that they are beyond help. In practice, such patients tend to routinely refuse hospitalization even when (or precisely when) there is abundant evidence that hospitalization is highly recommended.
On the other hand, patients who come to the ED requesting admission fall in one of the following categories: secondary gain, personality disorders, Axis I with enough insight to appropriately ask for help (not necessarily in this order). For obvious reasons hospitalization is not recommended for malingering and is rarely recommended for Axis II – as most personality disorders tend to flare up when excessive attention is poured over lingering character pathology, or Axis I with good insight – as insight tends to correlate with one’s ability to stay safe, ie, inversely correlates with need for hospitalization to preserve safety.
In other words, when a patient was brought to the ED by relatives, friends or police and states that nothing is wrong and there is no need for hospitalization, more times than not hospitalization is recommended. At the same time, when the patient brings himself to the ER requesting hospitalization, more times than not hospitalization is not recommended.
How do you cut this Gordian knot of in or out (of the hospital) when the patient is on the opposite side of your recommendation? Use your expertise to differentiate between false positive and false negative data. And, of essence, do not make up your mind before collecting as much collateral data as you can.