Pilot Assessment of Patient and Provider Characteristics Associated with Satisfactory Consultation-Liaison Telepsychiatry Encounters

Abstract

Background:

The COVID-19 pandemic created pressure to attempt remote consultation, but there are limited data on the use of telepsychiatry in general, and almost none about the experience of telepsychiatry in a consultation-liaison (CL) context.

Methods:

11 CL attending surveys and 8 attendings’ tele-encounter logs from March-June 2020 were completed and reviewed to assess for patient and provider characteristics associated with barriers to using telepsychiatry.

Results:

The vast majority of 223 tele-psychiatric encounters were acceptable to providers in terms of technology (82%) and their ability to form a connection with the patient (78%). In multivariable logistic regression models, an unresolvable difficulty utilizing the platform was less common for female patients (OR = 0.239, p = 0.002) and more common for patients who prefer a non-English language (OR = 9.059, p < 0.001); achieving a personal connection that felt right was also less likely for patients who prefer a non-English language (OR = 0.189, p = 0.001).

Conclusions:

Telepsychiatry has previously been limited to outpatient use and, generally, for providers and patients who specifically preferred it. However, abrupt transition to use of telepsychiatry to limit contagion risk was mostly satisfactory in our center; identifying for which patient encounters it is most and least appropriate will help guide future use.

Introduction

In-person care is the traditional standard for medical practice, but telemedicine and telepsychiatry have been increasingly employed, primarily to deliver care for outpatients too distant from providers to receive in-person care.¹ Telemedicine has been found feasible, reliable, acceptable and effective throughout a wide range of settings, specialties, age ranges, and cultural groups.^2,3,4,5 However, telepsychiatry literature is limited, heterogeneous, and largely focused on implementation process,^{6, 7} provider resistance,^{8, 9} and strategies to bolster physician acceptance.^{10, 11}

Considerations unique to the consultation-liaison (CL) setting may affect the feasibility, process, and acceptability of telepsychiatry, in explicit and indirect ways. An exclusively remote evaluation may obstruct diagnosis; patients may be too medically ill, fatigued, or delirious to attend to the videoconference. For inpatients, hospital personnel could potentially facilitate videoconferencing, but may experience this task as an imposition. Either the clinician or the patient engaged in remote work may encounter technological barriers or may feel less of a personal connection. There is minimal prior literature to guide implementation of in-hospital telepsychiatry consultation.^{5, 12, 13} For outpatients with medical comorbidity, similar concerns apply. As a group these patients are older, more frail, and more cognitively impaired than the general outpatient psychiatry population. However, they may also benefit most from avoiding an additional trip to our medical center.

During the COVID-19 pandemic, the need to reduce hospital traffic and in-person interaction heightened the importance of care delivered from a distance.^{12, 14} Telepsychiatry was abruptly implemented for consultations at our large urban CL program serving a tertiary care hospital. Patients and providers were therefore not self-selecting to use telepsychiatry as they had in the past.

Clinicians’ perception of a telepsychiatry program as beneficial, which predicts motivation to adapt, is integral to the success of a novel program.^{6, 10} Causes of physician resistance to telemedicine include a sense of reduced autonomy, situational anxiety and cost.⁸ Triage systems and workflows sometimes improve provider acceptance.^{5, 11} In this study we considered both provider and patient characteristics, hypothesizing that successful CL encounters – as defined by whether the signal was functional, whether the patient was able to successfully access the software and platform, and whether the provider felt the rapport built was adequate for the work -- would be associated with patient age, language, gender, diagnosis, and inpatient or outpatient setting.

Material and Methods

This study was approved by the Columbia University Medical Center Institutional Review Board. Physicians were recruited from our CL Division of the Department of Psychiatry requesting consent to use of ongoing QA logs and to participate in the survey. Logs of de-identified patient tele-encounters from March 2020 through June 2020 were included; providers completed surveys in June 2020.

Logged information included demographics, medical and psychiatric diagnosis categories, platform used, assistance in accessing teleconference, duration, billing codes, and barriers to care in these categories: (1) signal issues (none, any, or any but resolved) (2) individual user issues that impacted the patient’s ability to access the device, software, or platform (none, any, or any but resolved), and (3) personal rapport as determined by the treating psychiatrist (felt right, did not feel right, or could not assess due to issues better attributed to category #1 or #2).

A one-time identity-coded survey was also distributed to the physicians, querying basic demographics, access to relevant technologies at the time their higher education began, preferred device types and platforms, and general comfort with use of telepsychiatry.

Statistical Methods

We evaluated two binary outcomes: user issues and personal connection. For user issues, we merged responses into two categories: (1) no problem or problem that resolved and (2) problem that did not resolve. For personal connection, nine records were excluded due to user issues or signal issues, leaving two categories: (1) felt right and (2) did not feel right.

Patient characteristics including age, gender, preferred language, cognitive (delirium or dementia) versus non-cognitive (all other) psychiatric diagnoses, and inpatient versus outpatient setting were compared between different outcome statuses. There were so few signal issues that this variable was excluded from our final analysis. The sample of attendings was insufficient for statistical modeling but is included as descriptive statistics (Table A).

Table A –

Patient and Attending Details for Logged Tele-Encounters (n = 223)

PATIENT DETAILS
#encounters			(%)		#encounters	(%)
Gender				1° Med-Surg Disorder
	F	82	(36.77)	Cardiac	1	(0.45)
	M	123	(55.16)	Gastrointestinal/Hepatic	57	(25.56)
	O	18	(8.07)	Infectious Disease	40	(17.94)
				Nephrological	9	(4.04)
Age *				Neurological	6	(2.69)
Min=22	<30	32	(14.41)	Oncological	75	(33.63)
Max=82	30–39	20	(9.01)	Pulmonary	7	(3.14)
Mean=51.6	40–49	38	(17.12)	Other	28	(12.56)
SD=16.6	50–59	49	(22.07)
	60–69	57	(25.68)	1° Psychiatric Disorder
	70–79	19	(8.56)	Adjustment Disorder	26	(11.66)
	80–89	7	(3.15)	Anxiety Disorder	34	(15.25)
				Delirium	18	(8.07)
Location				Dementia	7	(3.14)
	Inpatient	118	(52.91)	Mood Disorder	72	(32.29)
	Outpatient	105	(47.09)	PTSD/Acute Stress	1	(0.45)
				Psychotic Disorder	11	(4.93)
Preferred Language				Somatoform Disorder	1	(0.45)
	Chinese	2	(0.9)	Substance Disorder	36	(16.14)
	English	164	(73.54)	Other	17	(7.62)
	German	7	(3.14)
	Greek	2	(0.9)	Setting/Location
	Spanish	48	(21.52)	Outpatient	105	(47.1)
				Inpatient	118	(52.9)
ATTENDING DETAILS
Age of attending involved in the encounter				Involved attending’s tech history**
	<40	144	(64.57)	Any	221	(99.1)
	40–49	76	(34.08)	Computer	160	(71.75)
	50–59	1	(0.45)	Internet	157	(70.4)
	60–69	2	(0.9)	Cell phone	44	(19.73)

^*n = 222 because of missing datum in one record

^**what kinds of tech they had access to in first year of undergraduate education

Variables associated with the outcomes with a p value < 0.1 in univariate analysis were included in the multi-variable logistic regression model. Firth’s bias reduced penalized likelihood logistic regression model was fitted to quantify the association between the outcome and covariates. Analyses were performed in Rstudio version1.1.453.

Results

15 CL attendings were invited to participate. 11 (73.3%) consented; 11 completed surveys; 9 submitted logs. No surveys were excluded. 268 encounters were collected, reflecting an unknown number of unique patients. After excluding entries not involving direct patient contact (23) or with incomplete outcome information (22), 223 entries of 8 attendings were included in the analysis. A total of 75 audio-only encounters and 139 videoconferenced encounters were included in our analyses.

Population Characteristics (Table A).

Encounters were 36.8% with female patients, 55.1% male, and 8.1% non-binary or other. The mean patient age was 51.6 years (SD 16.6). 52.9% of encounters were inpatient. 73.5% of encounters involved patients who preferred English, and 21.5% involved patients who preferred Spanish. A smaller number of encounters involved patients with a preferred language of German (3.1%), Greek (0.9%), or a dialect of Chinese (0.9%). Interpretation for all non-English languages was done via professional over-the-phone interpretation, with only a few encounters performed by providers who themselves were bilingual.

Logs were submitted by attendings whose ages ranged from 30s to 60s and whose undergraduate access to technology ranged from none to having a cellphone, internet access, and a computer. See Table A.

Analysis (Table B)

Table B – Multivariable logistic regression models for User Issues and Personal Connection.

Reference levels are Gender = male, Diagnosis = non-cognitive, Language = English, Setting/Location = outpatient, and Attending tech history is Internet = none and Cell = none

	OR	Lower 95%	Upper 95%	p value
Odds Ratio for having an unresolvable User Issues
Age	1.025*	0.990	1.064	0.159
Female gender	0.239	0.086	0.597	0.002
Non-binary/Other gender	0.403	0.003	4.894	0.528
Cognitive diagnosis	2.634	0.856	8.146	0.091
Preferred language is not English	9.059	3.397	27.027	<0.001
Inpatient setting	0.844	0.298	2.304	0.743
Odds Ratio for having a Personal Connection that “felt right”
Age	0.979**	0.945	1.012	0.211
Female gender	1.036	0.448	2.434	0.934
Non-binary/Other gender	0.460	0.031	68.307	0.660
Cognitive diagnosis	0.392	0.115	1.315	0.128
Preferred language is not English	0.189	0.060	0.528	0.001

^*for each year increase of age, the OR increases by 2.6%

^**for each year increase of age, the OR decreases by 2.1%

We had sufficient records to assess for the relationships between our independent variables and two outcomes: user issues, and rapport.

In the multivariable model for user issues, female patients were significantly less likely than male patients to encounter unresolvable user issues (OR = 0.239 [95% CI: 0.086, 0.597], p = 0.002), and patients preferring non-English languages were significantly more likely to have such issues (OR = 9.059 [95% CI: 3.397, 27.027], p < 0.001).

In the model for personal connection, a significant association was found between rapport and preferred language: patients who preferred any language other than English were less likely to have participated in an encounter with a positive assessment of rapport (for non-English: OR = 0.189 [95% CI: 0.060, 0.528], p =0.001). There was no relationship between diagnosis, age, or inpatient versus outpatient status and any of our index outcomes (Table B).

Therefore, in our population, after adjusting for all relevant variables, the significant findings were fewer user issues for female and English-preferring patients, and better rapport for English-preferring patients.

In a post-hoc Chi-square analysis, we also found that language predicted use of videoconference versus audio only, with English-preferring patients significantly more likely to be seen using a video visit (p < 0.001). However, there was no overall association between audio versus video visit and our outcomes of interest, either for the entire group, the English-preferring group, or the group of patients preferring other languages.

Discussion

In this study of 11 attendings surveyed and 8 attendings’ records of 223 telepsychiatry consultations, most encounters were acceptable in terms of signal, user issues, and rapport. Because there was little time for training or preparation, these data reflect pre-optimization telepsychiatry, suggesting that more time and training could make telepsychiatry even more effective. Some consultations may have succeeded because of less reliance on the tele-interaction and greater reliance on medical records or collateral information. In some cases, a failed attempt at tele-evaluation might have been replaced by an in-person consultation without recording the initial attempt.

For patients who preferred a non-English language, in spite of professional interpretation, we found a significantly higher number of unresolvable user issues and decreased rapport, consistent with general disparities in care for patients with limited English proficiency (LEP.)¹⁵ Patients who preferred languages other than English were also significantly more likely to have audio-only interactions, which we suspect was because interpreters were available via audio only, meaning that even when the psychiatrist was seen via video, the interpreter was not. Audio versus video did not predict assessment of rapport, which may reflect that the choice of audio or video was consistent with patients’ preferences.

Women had a significantly lower rate of user issues. Potential reasons include higher motivation, better skill utilizing assistance, or baseline superior technological competence. Age did not meet significance for predicting user issues, or rapport, suggesting that either age does not play a strong role in patients’ ability to access telepsychiatry, or that assistance from younger family members and hospital staff may compensate for age-related barriers. Diagnosis also did not correlate with user outcomes, which may reflect insufficient power to truly stratify by diagnosis. Inpatient versus outpatient location also did not meet significance for predicting user issues or rapport, in spite of inpatients being more ill. Very few patients in either location had internet signal issues. It is possible that socioeconomic differences might underlie a heterogeneity of experience in which some patient groups have better overall access from inside the hospital, where devices can be loaned, staff can assist, and internet speed is moderate, whereas others benefit from superior hardware and internet connection at home.

Strengths & Limitations

Logs were collected within a week of the COVID pandemic in a tertiary care center with extremely ill and diverse patients in the center of the first American wave of the pandemic.

Limitations and sources of possible bias include small sample size of providers, variable completeness of logs returned, unusually heightened motivation due to universal concern about contagion, an unknown total number of patients included in the anonymized logs, simple assessment options without nuanced questions interrogating how rapport was assessed, and lack of control groups including in-person comparisons and video language interpretation.

Future research should address these limitations by increasing sample size, utilizing Likert-type outcome scales, including patient perspectives, following long term rapport, and comparing video to audio interpretation for language-discordant encounters. Future work should also consider the use of consultation-liaison telepsychiatry specifically for critically ill patients and acute care settings, including ICU and surgical settings, and medical problems that limit hearing or vision. COVID both necessitated distant evaluation and changed in-person care in ways (high total patient burden, staff exhaustion, visitor restrictions, altered diagnostic make-up of the population) that could skew results, limiting generalizability. We did not have a control group to assess whether COVID-related barriers (copious personal protective equipment (PPE), shorter evaluations, lack of family presence) might have limited the quality of the assessment or rapport of in-person evaluations. We found no literature about the impact of provider face-coverings (for PPE or otherwise) on patient experience, although in our clinical experience patients often comment about the psychological impact of not seeing a provider’s face.

Conclusions

This pilot study showed that telepsychiatry can be feasible in a CL setting. We provisionally recommend that for services using a hybrid model, patients with LEP be given prioritized access to in-person consultation. Future research should focus on patient perspectives, include control groups, and consider the impact of provider facial coverings on rapport-building.