*3.3. BMI Change*

Table 3 summarizes clinical outcome parameters of AN patients on admission and at discharge from inpatient treatment. On average, AN patients achieved a weight gain of 2.1 ± 2.3 kg during the 32-day (25th percentile: 26 days; 75th percentile: 63 days) inpatient treatment program. The BMI increased by 0.7 ± 0.8 kg/m2, which is equivalent to a BMI increase of 4%. The mean rate of weight gain in AN was 0.29 kg/week and ranged from −0.44 kg/week up to 1.35 kg/week. Seven (14%) out of the 50 AN patients lost weight during their inpatient stay.


Data are expressed as mean ± SD (range) or as median (25th/75th percentile); BMI, Body Mass Index.

#### *3.4. Association between Physical Activity and Clinical Outcome*

In an univariate regression analysis with BMI increase in % as the dependent variable and a range of potential predictive factors as independent variables (length of inpatient stay, phase angle, BMI on admission, steps, total distance, PA at different MET intensities, duration of sleep and recumbency), only length of inpatient stay (*r* = 0.154; *p* < 0.001), phase angle (*r* = −2.95; *p* = 0.002) and BMI on admission (*r* = −1.99; *p* < 0.001) were significant predictors (presented in Figure 1 with Spearman rank correlation). In a multivariable model, length of inpatient stay (*p* < 0.001) and BMI on admission (*p* = 0.029) remained significant predictors and duration of sleep became significant (*r* = −0.0107; *p* = 0.019) as well. In addition, for MPA a trend (*r* = 0.0111; *p* = 0.089) towards becoming a significant positive predictor of BMI increase was observed.

**Figure 1.** Associations between BMI increase in % and (**a**) length of inpatient stay and (**b**) number of steps per day applying Spearman rank correlation. BMI, Body Mass Index.

In an exploratory regression tree model (Figure 2), the following parameters were relevant predictors of percent BMI change: length of inpatient stay, BMI on admission, and number of steps.

**Figure 2.** Regression tree for non-linear modelling to test the relation between BMI percent change and further parameters. BMI, Body Mass Index.

With this non-linear model, the association between actual and predicted BMI percent change could be predicted with an *r*<sup>2</sup> = 0.81 (Figure 3).

**Figure 3.** Non-linear model on predicted vs. measured BMI change. BMI, Body Mass Index.

Furthermore, validation of this prediction model by jack-knife analysis was successful (*r*<sup>2</sup> = 0.203; *p* < 0.001). The importance score for length of inpatient stay was 1026, for admission BMI 822, and for number of steps 453, potentially suggesting at least a small effect of PA measured as steps on % BMI increase. Applying these statistical procedures also for LPA as a parameter for low level activity, data yielded a similar value for *r*<sup>2</sup> for steps as well as an importance score of 562 for LPA, indicating also a slight effect of LPA on BMI change comparable in strength to that of steps.

#### **4. Discussion**

In our study, the following main results emerged: (1) Compared with healthy controls, AN patients spent more time engaging in light and less time engaging in vigorous intensity PA; (2) the patient's BMI increase during inpatient treatment was largely predicted by low BMI on admission and longer duration of inpatient stay; (3) high step count and time in light-intensity PA only emerged as potential predictors of lower BMI increase in an exploratory and non-linear model; and (4) contrary to our assumption, the duration of sleep as a marker of physical inactivity was inversely associated with BMI increase.

Few studies have objectively measured low intensity PA in AN inpatients compared with healthy controls. Our findings demonstrating increased low intensity PA are consistent with a previous study using a shoe-based monitor [11]. Using the SenseWear armband, El Ghoch et al. [3] also observed that AN patients spent less time in high intensity PA, yet contrary to our findings time spent in light-intensity PA did not differ between groups and the patients showed a significantly higher moderate and vigorous PA duration. The division into two low intensity MET categories (1.1–1.8 and 1.8–3) in the present study as opposed to one category (1.1–3 METs) in the study by El Ghoch [3] may explain the different and more detailed results. Differences in time spent with moderate to vigorous PA might relate to varying approaches with respect to the restriction of PA on the wards or to practical opportunities to exercise in the environment outside the ward. The choice of different PA assessment tools should also be taken into account: When assessing PA with movement sensors, there was no difference in time spent on "fidgeting" (operationalized as "body position change counts") between AN patients and controls [8]. Yet, the authors of that study mentioned problems with measurement technology consisting of several leads and wires, which might have affected compliance and PA behavior of the study participants.

A better understanding of the origin of distinct PA patterns in AN patients is warranted. Increased light PA in AN might be a consequence of negative energy balance resulting in a foraging response to increase PA to find food [19,20] or linked with a distinct phenotype characterized by disturbed energy homeostasis specifically associated with increased PA despite severe weight loss [6,21]. Further origins for increased light PA in AN might be an attempt of emotion regulation [6,22] or the desire to lose weight [4,23]. Interestingly, when interviewed 57 years after participating in the Minnesota starvation experiment in 1944/1945, the volunteer men did not report an increased drive for PA while starving [23]. To add to the complexity of altered PA in AN, the surrounding conditions during the time of PA assessment might play a considerable role. When obese volunteers were subjected to 24-h measurements of energy turnover within a metabolic chamber [24] for analysis of spontaneous PA, with exercise being prohibited within the chamber, the authors hypothesized that such forced reduction of voluntary exercise may have resulted in the partially observed increased engagement in spontaneous PA [25]. Similarly, in 16 healthy, male volunteers who underwent 8 weeks of experimental overfeeding, two-thirds of the increase in total daily energy expenditure was due to increased non-exercise activity thermogenesis (NEAT) [26]. Individual variation in NEAT accounted for the 10-fold differences in fat storage that occurred with overfeeding, suggesting that during positive energy balance, high activation of NEAT results in difficulties to gain weight for some individuals. The phenomenon of high NEAT and concomitantly energy needs of 4000 kcal/day to gain weight was recently documented in a case report of a young woman with AN at the end of therapy [27]. We believe that it is important for the tailoring of suitable PA interventions for AN patients to find an answer to the question whether increased low-level PA is an AN-specific phenotype that is linked with physiological processes during starvation and refeeding, or whether such PA behavior is related to restrictive treatment setting characteristics irrespective of AN, which may also be observable in other populations. Therefore, the current restrictive handling of PA during AN treatment may need to be reconsidered since an increase in low-level PA could provoke higher daily energy expenditure and might hinder weight recovery. Increased voluntary exercise could be accompanied by a decrease in spontaneous PA [25]. Concomitantly, Calogero et al. [28] investigated the effectiveness of an exercise program in patients with eating disorders, reporting on weight improvements through this intervention and concluding that patients in the exercise program may have been less likely to exercise secretly, whereas patients in the control group may have exercised unsupervised.

A low admission BMI was identified as a major determinant for BMI increase in AN. Resting energy expenditure proportionally declines with BMI [29], physiologically leading to a more rapid weight regain at the beginning of treatment. Longer inpatient stay also predicted BMI increase which may be at least in part explained by the rules in our adult treatment setting where patients were discharged if they continuously failed to meet the expected weight targets. On the other hand, patients who stayed in treatment longer also had more time to gain weight. Since only a slight effect of PA on BMI increase was observed and only by conducting an exploratory analysis, the admission BMI and duration of stay had an overall much greater, independent and overriding predictive power on BMI increase in AN. Whether an association between PA and weight trajectory in AN can be detected may depend on certain study characteristics, i.e., measurement technique and time point of PA assessment. There was no association between (I) PA duration at different MET intensities and daily steps at inpatient discharge and BMI at 1-year follow up [30]; (II) PA level operationalized as the average acceleration in m/s2/min from both feet and BMI or rate of weight gain in AN patients admitted to an inpatient unit [11], and (III) time spent on feet at low-weight within 2 weeks of hospital admission or 1-month post-treatment discharge and 12 months BMI trajectory [8]. However, a longer on-feet duration at the inpatient weight restored time point was associated with a more rapid decrease in BMI over the 12 months following discharge [8]. Interestingly, a retrospective study applying questionnaires for PA assessment 6 months and 1 week prior to inpatient admission in 20 adolescents with AN found that an increase in PA—and not a decrease in food intake—was associated with the need for inpatient treatment [31]. These findings give rise to at least some effect of PA on the weight and illness trajectory of AN, and stress the need for further, systematic studies on this topic.

In the present study, no linear and direct associations between sleep duration and recumbency, conceptualized as physical inactivity parameters, and weight gain during AN treatment existed. Similarly, others could not find associations between sleep patterns and BMI [32], or between changes in sleep patterns and changes in BMI [33]. In the latter study, there was a significant direct association between baseline sleep time and BMI. In the present study, the contribution of sleep duration to variance of BMI increase was only of minor effect size. However, unexpectedly, in a multivariable model, the duration of sleep was inversely associated with BMI increase. In obesity, short sleep duration is known to be associated with increased food intake and excess body weight [34,35]. Whether this link also applies to patients with AN needs further investigation.

While we used objective PA assessments at standardized time points close to hospital admission, which are different from other studies that assessed PA across various stages of AN treatment [11], our findings also need to be interpreted within their limitations. Firstly, the validity of the SenseWear armband in severely underweight AN patients is unknown, and raw data and algorithms within the armband software are not accessible to researchers. Nonetheless, we consider this technology suitable for PA detection due to its easy handling compared to other devices [8] and the fact that multiple sensors enable the distinction between various types of PA, the recording of actual on-body time as well as time spent on sleep [14]. Second, whether our controls were of comparable socioeconomic background, and whether the wearing of the armband motivated them to work out more than usual remains unclear. Given that PA analysis was conducted between hospitalized AN patients and healthy controls within their everyday environment and thus in two very different settings, comparability of data may be argued. As a consequence, PA patterns of the patients in the present study may not be representative of other patients with AN under other types of care. However, finding a suitable control group for hospitalized patients is difficult, as healthy people are not hospitalized, and hospitalized patients for other reasons than AN are likely to suffer from a medical condition which affects PA patterns.

In conclusion, we found that AN patients spent more time engaging in light, and less time engaging in vigorous intensity PA than controls, and that the BMI increase during inpatient treatment was predicted by low admission BMI and longer inpatient treatment. Furthermore, high step count and time in light-intensity PA only emerged as potential predictors of lower BMI increase in an exploratory and non-linear model. This latter finding indicates that the effect of PA on the disease course of AN should be quantified and clarified further and that more complex models may need to be employed in future research on this topic. Since PA behavior is likely influenced by multiple factors including age, psychological and nutritional parameters, assessment of these potential modifiers in future studies may contribute to a better understanding of PA variability in AN.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2077-0383/7/9/289/s1, Table S1: Medical details, comorbidities, and medication of study patients and the healthy controls.

**Author Contributions:** Data curation, C.S.L., T.H. and A.S.; Project administration, T.H., U.E., A.S. and V.H.; Writing—original draft, C.S.L.; Writing—review & editing, T.H., U.E., M.R., C.U.C., A.S. and V.H.

**Funding:** This work was supported by funding of the Swiss Anorexia Nervosa Foundation (Project Number 23-13) and Charité University Funding (UFF 89/441-176, A.S.).

**Acknowledgments:** We thank Andreas Busjahn for his support with the statistical analysis; Magdalena Brinkmann, Karin Johansson and Christina Hentzschel for their assistance with the organization of clinical assessments. We acknowledge support by Deutsche Forschungsgemeinschaft and Open Access Publishing Fund of Charité University Berlin.

**Conflicts of Interest:** Correll has been a consultant and/or advisor to or has received honoraria from: Alkermes, Allergan, Angelini, Gerson Lehrman Group, IntraCellular Therapies, Janssen/J&J, LB Pharma, Lundbeck, Medavante, Medscape, Merck, Neurocrine, Otsuka, Pfizer, ROVI, Servier, Sunovion, Takeda, and Teva. He has provided expert testimony for Bristol-Myers Squibb, Janssen, and Otsuka. He served on a Data Safety Monitoring Board for Lundbeck, ROVI and Teva. He received royalties from UpToDate and grant support from Janssen and Takeda. He is also a shareholder of LB Pharma.
