Use of Security Officers on Inpatient Psychiatry Units

The New York State Psychiatric Institute Institutional Review Board approved this study. Data were collected at Allen Hospital, a 200-bed community hospital in New York City that provides inpatient and outpatient medical and surgical services. The hospital serves a multiethnic population. Revenue primarily comes from billing Medicaid, Medicare, and private insurance companies. At the time of this study, the 30-bed locked inpatient psychiatry unit was staffed to treat men and women with co-occurring mental illness and substance use disorders. Patients were admitted from the emergency room and the medical and surgical floors. The unit was always at capacity and never exceeded capacity.

Allen Hospital has security officers stationed on a floor other than the floor for the inpatient psychiatry unit. Officers conducted periodic rounds on the inpatient unit. In addition, any staff member could request security assistance via telephone or panic alarm. When called, two to six uniformed security officers arrived within minutes to assist clinical staff. Security officers are unarmed.

Several hospital policies govern the use of security officers. Security officers can be utilized for “imminent risk of violence” (“a situation characterized by aggressive or potentially aggressive behavior or threats by one or more patients”). Security personnel do not provide clinical services (“Security personnel may be utilized for safety purposes only, never for clinical observation . . . [officers] provide no clinical services”). Security officers work under the direction of clinical staff (“Based on the threat level and the potential for violence, a security officer may be assigned to act as safety back-up to assist the clinical staff with a particular patient”). Security officers are trained in crisis intervention techniques by members of both the security and the psychiatric departments (“All security personnel receive annual training in crisis management techniques for psychiatric emergencies”) (New York–Presbyterian Hospital Policy and Procedure Manual, 2016, unpublished).

First, patient charts were reviewed to identify all inpatients discharged from July to December 2016. No major changes in staffing or floor management occurred during that period. One author (REL) completed a data sheet for each discharge.

Demographic and clinical data were collected for each patient, with a focus on variables potentially associated with violence and aggression. These included demographic factors (age [9,10,15–17], sex [7,15,16,18], and homelessness [19]), historical factors (multiple prior hospitalizations [9,16], prior state hospitalizations, history of violence [8–10,16,17,20], and history of substance abuse [7,9,15,16,20]), current clinical factors (psychosis [4,9,15,16,18,20], dementia [15], medication nonadherence before admission [21], receipt of intramuscular antipsychotics in the emergency room [22], and receipt of assertive community treatment [ACT] or assisted outpatient treatment [AOT] at the time of presentation), and legal factors (involuntary hospitalization [7,15,16], receipt of a new court order for AOT, and going to court for any reason). Going to court for any reason refers to hearings where a judge was asked for a ruling on treatment over objection, involuntary retention in the hospital, extension of an involuntary hospitalization beyond 60 days, or AOT. The exact wording of each variable is shown in Table 1.

Second, the security logbook was reviewed to identify events when security officers were dispatched to the inpatient psychiatry unit for urgent or emergency situations while these patients were in the hospital. The security logbook is kept in the security office and lists the date of each call, the patient’s name and medical record number, and a brief description of what happened (for example, “patient was medicated by nurse”). One of the authors (REL) reviewed the logbook to identify security calls to the inpatient unit. False alarms and security escorts during patient transfers (between hospital units or to court) were not included in the analysis. Per hospital policy, a security officer always accompanies a patient to court; these routine escorts were not included in the analysis, to keep the focus on urgent calls to the inpatient unit.

Third, events from the security logbook were linked with descriptions in the medical record (physician notes, nursing notes, and medications). Two authors (REL and MPC) used a violence scale to code patient behaviors associated with the security call (one code per call) (23). Behavior codes were as follows: no violence, loss of impulse control (behavior described by such words as “running wildly,” “yelling,” “shouting,” and so forth and giving no indication of this activity’s being directed at persons or objects), ambiguous violence (nonspecific description of behavior as violent or aggressive without elaboration), verbal attacks on persons (verbal behavior that is not explicitly threatening but is pointedly abusive and hostile), threats to persons (acts or words that convey the possibility of inflicting harm), attacks on objects (throwing, breaking, or hitting nonhuman objects), attacks on persons (physical assaults), or unknown (not clearly documented). These authors (REL and MPC) coded interventions as follows: intramuscular injection involving an antipsychotic, intramuscular injection not involving an antipsychotic, oral medication involving an antipsychotic, oral medication not involving an antipsychotic, other intervention, or unknown (not clearly documented).

These authors also assigned a confidence level code (0, 1, or 2) each time they tried to match an incident reported in the security logbook to an event reported in the medical record (0, security logbook reported an event but one or both reviewers did not think the medical record described a corresponding event; 1, security logbook and medical record both reported an event but the medical record did not explicitly mention security being present; 2, security logbook and medical record both described an event and the medical record explicitly mentioned security being present). Two authors (REL and MPC) assigned codes at the same time and with the same computer, and differences were discussed until consensus was reached.

An author (REL) reviewed the nursing logbook, which records incidents of seclusion, mechanical restraint, and manual holds.

The first analysis used security calls as the unit of analysis. Simple counts were generated of when security calls happened (day of the week and hospitalization day), number of calls per patient, patient behaviors associated with security calls, and clinical interventions.

The second analysis used the individual patient as the unit of analysis and assessed which patient characteristics were associated with security calls. For patients who had more than one hospitalization during the study period, only the first hospitalization was included in the analysis (nine patients had two hospitalizations, and one patient had three hospitalizations). Security calls were treated as a binary variable (no security calls versus one or more security calls).

Bivariate models (chi-square and t tests) were constructed to identify associations between each patient variable and security calls. Variables with significant bivariate associations (p<.05) were included in a stepwise logistic regression model to determine whether associations persisted after adjustment for other relevant variables.

Statistical analyses were performed with Stata/IC 12.1 for Windows.

The sample included 272 unique patients, described in Table 1. There were 161 security calls. Three calls listed in the security logbook did not specify a patient, and three calls in the medical record were not listed in the security logbook. However, the dates did not match, so they were treated as separate calls (all three extra calls found in the medical record were for patients who already had multiple security calls).

One patient was an outlier, with 50 security calls. This patient refused to take medication despite a court order, and nursing called security once or twice each day—as a precautionary measure—while administering medication over objection. In the presence of security, the patient accepted medication without physical resistance. This patient was removed from the security calls analyses because there was risk of skewing the data.

After removing the outlier, analyses indicated that security calls did not significantly vary by day of the week (Figure 1). Calls were most common during the first week of hospitalization. Of the patients who experienced a security call, half had only one call.

Ninety-four security calls could be confidently linked (confidence levels 1 and 2, outlier excluded) to behaviors and interventions described in patients’ medical records. The most common behaviors associated with security calls were “threats to persons” (N=34 calls; 20 threats to staff, eight threats to patients, three threats to both, and three nonspecific threats to harm someone) (Figure 2). “Attacks on persons” were associated with seven security calls. The most common intervention was an intramuscular antipsychotic injection (N=49 calls), followed by oral antipsychotic medication (N=29). A nonmedication intervention occurred for eight calls: a visitor was escorted out (N=3), the patient was escorted to his or her room, the psychiatrist talked to the patient, security talked to the patient, a room search for contraband was conducted, and in one instance the patient had just received an injection. The intervention was not documented in two instances.

For the 17 security calls that could not be confidently linked to the medical record (confidence level 0, outlier excluded), it was not possible to ascertain patient behaviors associated with the security call. However, the security logbook reported interventions in all but one case: the patient was medicated (N=10), the patient was spoken to (N=5), and a visitor was dismissed (N=1).

The nursing log recorded no episodes of seclusion or mechanical restraint during the study period. On four occasions, a patient was briefly held so that medication could be safely administered (holds lasted seconds to minutes).

Most patients (N=223) had no security calls; 49 patients (18%) generated 157 security calls (three additional security calls were missing data and could not be linked to a unique patient, and one security call was excluded because it was for a repeat hospitalization). Security calls were associated with a longer hospital stay (from the day the person entered the emergency room until final discharge from the hospital: 33.0±23.3 days versus 19.5±15.0 days for patients with no security calls, t=−5.09, df=270, p<.001).

In bivariate analyses, security calls were associated with several historical factors (more than one prior hospitalization, state hospitalization, and violence history), current clinical factors (medication nonadherence contributing to the hospitalization, receipt of an intramuscular antipsychotic in the emergency room, and current receipt of ACT or AOT), and legal factors (involuntary hospitalization, requiring a new AOT order, and going to court for any reason) (Table 2).

In the stepwise logistic regression model, security calls were significantly associated with three patient characteristics: having more than one prior hospitalization, involuntary hospitalization, and going to court for any reason (Table 3).